ANALYSIS OF GLACIER MASS BALANCE AND RHEOLOGY AT KEKESAYI GLACIER USING HEXAGON KH-9, ALOS-PRISM AND SAR DATA

Transcription

1 ANALYSIS OF GLACIER MASS BALANCE AND RHEOLOGY AT KEKESAYI GLACIER USING HEXAGON KH-9, ALOS-PRISM AND SAR DATA Nicolai Holzer (1), Saurabh Vijay (1, 2), Manfred Buchroithner (1), Manoj K. Arora (2) (1) Institute of Cartography, Technische Universität Dresden, Dresden, Germany. (2) Geomatics Engineering Group, Department of Civil Engineering, Indian Institute of Technology Roorkee, Uttarakhand, India. ABSTRACT Scanned stereoscopic spying photography of a Hexagon KH-9 mission in 1973 as well as high resolution stereo imagery of ALOS-PRISM from 2009 was used to generate multi-temporal Digital Terrain Models for the estimation of glacier thickness changes of Kekesayi glacier situated at the Mustag Ata massif at the northwestern end of the Tibetan Plateau. Glacier surface velocities were estimated by amplitude tracking techniques based on a TerraSAR-X data pair acquired in Preliminary results show ice thickness loss for the ablation area of this glacier of about -20 meters for the time period between 1973 and Surface velocities are reaching up to cm per day at the upper part of the glacier in 2011, lowering significantly down in direction to the glacier tongue. Key words: Hexagon KH-9; ALOS-PRISM; TerraSAR- X; Digital Terrain Model; Glacier mass balance; Glacier surface velocity; Tibet; Pamir 1. INTRODUCTION Glaciers as sensitive climate indicators are of a high interest for climate change studies worldwide, and play particularly in arid regions such as Mustag Ata an important regulating role in regard of freshwater supply. Kekesayi glacier is a debris covered valley glacier which is originating from the eastern slope of Mustag Ata (7546 m) at the north-western edge of the Tibetan Plateau, situated in the Xinjiang region of China. With a length of approximately 18 km it is the largest glacier at this massif. The Mustag Ata / Kongur Shan mountain range is the most western benchmark region that is studied as part of the research project WET (Variability and trends in water balance components on the Tibetan Plateau) which aims to further understand the coupling of climate and hydrological cycles on the Tibetan Plateau. 2. STUDY AREA The cold and arid climate of this region is principally influenced by the mid-latitude westerlies, and still under minor influence of the Indian monsoon. With a mean annual precipitation to the glacier accumulation zone at Mustag Ata of about 300 mm at 5910 m asl. and with highest precipitation occurring in spring, it is suggested that the continental type valley glaciers in this region are more sensitive to a change in precipitation as to temperature [2, 4]. It could be proved that the glaciers in the Mustag Ata and Kongur Shan mountains decreased by 7.9% in area and by 9.9 % in glacier frontal position for the time period of 1962/66 to 1999 [5]. In another study, surface velocity values of in average 3.4 m for 44 days for the upper area of Kekesayi glacier could be determined by the use of ALOS PALSAR data collected in 2009, with much slower velocities glacier downstream [9]. Positive mass balance for an undefined glacier in the Mustag Ata massif has been reported for four of five consecutive observations years in between 2006 and 2010 based on in-situ measurements [10]. 3. DATA Research based on multi-temporal remote sensing data which is reaching back until the 1960s is possible by taking advantage of declassified high resolution stereoscopic imagery of the US Keyhole (KH) Corona (KH-4a, KH-4b) and Hexagon (KH-9) missions. The datasets of these spy missions which provided invaluable information for the US American government during the Cold War area offer stereocapability and were declassified in 1995 for Corona imagery and in 2002 for Hexagon data of the mapping camera respectively. Large ground coverage of the Corona imagery ( ) could be realized by mounting a panoramic frame camera on the satellite, which however resulted in complex spatial distortion patterns of the imagery. The newer dating imagery of the KH-9 mission ( ) is free of such spatial panoramic distortions due to a modified and improved camera system. Moreover, it contains a reseau grid in the imagery that allows reconstructing the image geometry conditions when to photography was taken at the time of exposure. However, most of the KH-9 related documentation still remains classified by now, making it difficult to process the data. The film size of a Hexagon KH-9 photography recorded by its mapping camera is 46 x 23 cm by having a

2 resolution of 6-9 m GSD (scanned at 14µm) which covers an area of about 250 to 125 km on the ground. The images contain four fiducial marks and 1058 reseau marks, and allow with its 70% overlap triplet stereo coverage to generate DTMs also of mountainous areas [7]. Actual high resolution optical remote sensing imagery with stereo capability includes ALOS-PRISM, SPOT-5, and Cartosat-1. For our benchmark region Mustag Ata we purchased ALOS imagery both of the panchromatic PRISM sensor with its stereo imagery (2.5 m GSD) as well as of the multispectral AVNIR-2 sensor (10 m GSD) due to good could and ice free data availability at about the end of the ablation period (September 2009). ALOS-PRISM imagery offers with its forward, nadir and backward looking camera triplet coverage of a scene, which allows obtaining good DTM quality in also high mountain areas. A disadvantage is the limited radiometric resolution of only 8 bit, what results in very poor contrast particularly in areas of snow and ice. In contradiction to optical imagery, synthetic aperture radar (SAR) data such as from TerraSAR-X consists of useful amplitude and phase information of the Earth surface target and hence both can be exploited to measure the subtle changes over Earth surface using numerical methods viz. DInSAR and Amplitude Tracking methods respectively. Such techniques may also allow separating active from non-active glacier areas lying under debris cover. 4. METHODOLOGY Good contrast in stereo imageries is essential in order to extract Digital Terrain Models out of such data. For glacier covered areas this might become problematic in case of snow and ice-cover representing poor texture in an image, sometimes even in case of high geometric and radiometric resolution. For debris-covered glaciers however, such as Kekesayi glacier, the contrast in optical images is at least for the ablation zone - in general much better, making such glacier types suitable for geodetic mass balance determination methods based on optical data. The estimation of ice thickness changes and the (to-be) subsequent determination of mass balance of Kekesayi glacier is based on two multi-temporal Digital Terrain Models (DTMs) generated out of high resolution stereoscopic remote sensing imagery recorded in 1973 and The glacier surface elevations representing the situation in August 1973 are obtained from a DTM derived out of declassified scanned stereoscopic spying photographs recorded from the Hexagon KH-9 (Keyhole-9) mapping camera. A DTM of the actual date was extracted out of high resolution triplet-coverage stereo imagery of ALOS-PRISM in September For the estimation of surface velocities of Kekesayi glacier, amplitude tracking techniques by using TerraSAR-X stripmap mode data with 11 days temporal separation acquired in August 2011 were exploited (Tab. 1). Table 1. TerraSAR-X data and its date of acquisition used in this study Parameters TSX-1 TSX-2 TSX-3 Date of acquisition 10/08/ /08/ /09/ Image processing for multi-temporal DTM extraction and geodetic mass balance For image processing and DTM extraction of the scanned Hexagon KH-9 photography we followed a method that was already successfully employed in previous studies [3, 7]. We tried to reconstruct the original image geometry by correcting the film distortion of the photograph that evolved over time because of film duplication and the almost four decades of storage. To obtain the imagery under the condition as it was taken at the time of exposure in 1973 we can take advantage of reseau grid coordinates in the original KH-9 photograph. For doing so, we developed a program in the Python programming language for automating this process. First, all actual reseau grid positions in the image are detected and the theoretic reseau grid positions (1 cm spacing) are calculated. Secondly, the reseau grids in the imagery are eliminated via bicubic interpolation by the use of surrounding pixels, since these reseau marks would disturb the later DTM extraction process. Finally, based on the known coordinates of the actual and theoretic reseau grid positions, the film distortion is corrected with polynomial image correction of a second order. Wallis filtering and histogram enhancement is finally employed to enhance the contrast. Unfortunately, fiducial marks were not scanned in all images correctly, so we assumed the centre of the reseau grid as principle point. Distortion vectors from the actual to the theoretic reseau marks do also show a rotation component (here with the principle point in the centre) that originates from an occasionally slight rotated scan of the films (Fig. 1). The inner orientation of the KH-9 mapping camera is unknown, but it is assumed to be similar to the NASA Large Format Camera (LFC) of 1984 (30.5 cm focal length) [7]. Since no information is known about ephemeris data to the authors, we obtain the exterior orientation only from Ground Control Points (GCPs) measured in the panchromatic band (14.25 m GSD) of orthorectified Landsat 7 ETM+ imagery for X and Y, and of SRTM3 version 4 elevation data for Z, what is very time consuming. The moderate accuracy that can be obtained out of such GCPs is probably the major influence factor

3 in regard of the accuracy of the later extracted DTM and subsequently the geodetic mass balance determination. However, due to the remoteness of the investigation area as well as due to political restrictions, it is not possible to obtain more accurate ground truth data, such as from in-situ GPS measurements. Anthropogenic objects such as crossroads of streets are ideal for GCP measurements. Another issue in this regard is the fact that due to the long temporal baseline of nearly 40 years such actual objects did often not exist back in 1973, or they are not visible the lower resolution imagery of Hexagon. Subsequently, natural objects such as river meetings have often to be determined as GCPs. Figure 1. Distortion vectors from actual to theoretic reseau grid coordinates of a Hexagon KH-9 imagery covering Mustag Ata (scanned in two parts) Based on this approach, we succeeded in extracting a preliminary Digital Terrain Model out of the previously processed triplet coverage Hexagon KH-9 photographs with a resolution of 30 m, based on 20 measured GCPs on flat and glacier free areas. The triangulation RSME of the bundle block adjustment is 0.81 pixels. Based on the purchased ALOS-PRISM imagery by using only its Rational Polynomial Coefficients (RPCs) provided for exterior orientation, we extracted a DTM of 10 m resolution. The triangulation RSME of the bundle block adjustment resulted here in a value of 0.28 pixels. Extracting a DTM by also taking advantage of additionally measured six GCPs from Landsat 7 ETM+ and SRTM3 shows worse results, probably due to their poorer accuracy in relation to the high resolution of ALOS PRISM and its RPCs. It turned out that the contrast of the ALOS-PRISM imagery must still be enhanced because of unsatisfactory DTM results in several low contrast areas a problem apparently related to the low radiometric resolution of ALOS- PRISM Techniques for estimation of glacier surface velocities Differential SAR Interferometry (2-pass DInSAR) is the extension of InSAR processing to estimate elusive changes over the Earth surface which was first used in 1993 in glaciological context [1]. The underlying concept of DInSAR is to analyze the phase difference of the target in coregistered complex SAR image pairs i.e. master-slave images after removing phase corresponding to topography, atmospheric influences, flat earth and other noises. The TSX-1 and TSX-2 SAR image pair is processed to calculate interferometric phase and coherence as shown in Fig 3. The coherence is a fundamental parameter to measure the quality of interferometric data and it is a function of decorrelation [8, 11]. In order to solve the phase ambiguity caused by the fact that absolute phase is wrapped into the interval (-π, π), phase unwrapping is done and finally the metric displacement in slant range direction (satellite imaging direction) can be estimated. SAR offset-tracking method based on amplitude or intensity can be used as an alternative to DInSAR which could even be employed when there is no coherence [6]. It has the other advantage of estimating displacement in radar flight direction (azimuth). The underlying concept of Amplitude Tracking is that it calculates pixel offsets in azimuth and range direction corresponding to maximum value of cross-correlation function computed over reference block (master image) and search block (slave image).

4 5. PRELIMINARY RESULTS AND DISCUSSION 5.1. Determination of ice thickness changes By comparing both generated DTMs of the year 1973 and of 2009, our preliminary results show negative glacier thickness changes of about -20 m of loss in mean for the ablation area of Kekesayi glacier (Fig. 2). The vertical accuracy of the KH-9 Hexagon DTM of 1973 compared with 20 measured ground control points from Landsat 7 ETM+ and SRTM3 used to extract this DTM currently still result in a mean error of 8.4 m with an RMSE of 24.5 m. For the DTM of 2009 extracted out of ALOS PRISM imagery by using its RPC data, the mean error to six measured ground control points from Landsat 7 ETM+ and SRTM3 is -2.9 m with an RMSE of 5.0 m. Both DTMs of 1973 and 2009 compared with ICESat GLA14 altimetry data show acceptable results of ± 10 m for non-glaciated and flat (slope < 15 ) areas, in case of a good quality in the DTMs for regions with sufficient contrast in the imageries Estimates of glacier surface velocities In Fig. 3, the interferogram generated from TSX-1 and TSX-2 show complete decorrelation over the entire glacier surface and fringe density is more over stable high mountain region nearby glacier. The low coherence value (black) over glacier surface justifies the temporal decorrelation. Figure 3. InSAR processing of 10/08/2011 and 21/08/2011 gives a) Amplitude Image of 10/08/2011, b) Multilooked (2 x 5 in range and azimuth) interferogram generated using this interferometric pair and Goldstein filter which is applied to remove the phase noise and enhance the contrast of the fringe, showing the complete decorrelation over glacier surface. C) Coherence image. Figure 2. Difference image (DTM of 1973 minus DTM of 2009) and profile taken glacier upwards in the middle of the glacier. Preliminary results show a glacier thickness loss of -20 m in mean for the ablation area due to positive values in the difference image. In Fig. 4, the two dimensional (azimuth and range plane) velocity map of the Kekesayi glacier surface is shown and in addition four profiles (Profile 1-4) and three hypothetical sections (A, B and C) are located to illustrate the detailed surface velocity at the different parts of glacier stream. From the profile results it appears that the upper part of the Kekesayi Glacier, which is located at the east side of Muztag Ata massif peak, has the velocity ranging around cm / day

5 and the velocity becomes 8-16 cm / day where the glacier bends to an north-eastern aspect. The downstream region of the glacier moves with very low velocity of around 2 6 cm / day. It s also quite evident that glacier velocity nearby stable areas are more than its central longitudinal profile because of frictional forces. Figure 4. The two dimensional velocity map of Kekesayi glacier for interferometric pair (master: 10/08/2011 and slave: 21/08/2011) showing the location of four profiles (Profile 1 to 4) and three hypothetical sections A, B and C (Profile 1: upper left; Profile 2: upper right; Profile 3: lower left; Profile 4: lower right) The accuracy of the Amplitude Tracking could be estimated by considering the pixel spacing value in the two governing direction (azimuth and range) and the oversampling factor used for sub-pixel image registration [6]. The expected error is supposed to be 5.63 cm and cm in slant range and azimuth direction with pixel spacing of 0.9 m and 3.03 m respectively and an oversampling rate of 1/16 th. By also considering the incidence angle, the expected error in ground range direction is cm. For the temporal baseline of 11 days, the error will be corresponding to ~0.74 cm/day and ~ 1.7 cm/day. In the azimuth-range plane the error would be 1.85 cm/day. The slope existing above glacier of around 1500 m x 1000 m is used where the displacement velocity can be convincingly assumed to be zero. The mean displacement velocity of this area comes out to be 2.46 cm/day. The estimated accuracy is very promising as compared to the expected one, however the accuracy should be preferably measured using in-situ observations, which are not available for this study. No interferometric fringes and very low coherence are observed over the glacier surface. This signifies that phase based algorithm, DInSAR, fail over the fast moving glacier surface. In simple words, movement between image acquisitions is more than the half of the wavelength of the signal i.e. more than 1.55 cm over period of 11 days in case of TSX imagery. So it can be concluded that interferometric phase is not well retained due to temporal decorrelation to estimate surface velocity of Kekesayi glacier using TerraSAR-X data and Amplitude Tracking is a good alternate. 6. OUTLOOK Despite of the still moderate quality of the current DTM derived from scanned Hexagon KH-9 imagery of its mapping camera, we believe that the datasets of the long-term US Keyhole (KH) spy program which are dating back until the 1960s provide very valuable data not only for example in regard of change studies of land cover and land use, but also for the cryosphere community. We expect a higher accurate DTM after absolute and relative vertical accuracy improvements by taking also advantage of ICESat altimetry data and by a better co-registration of both DTMs, in order to make more reliable conclusions in terms of mass balance change. Finally, we need to merge the results of glacier mass balance with the ones of glacier velocity measurements to make further conclusions in regard of glacier behaviour. Further very high resolution imagery of the Gambit-3 KH-8 mission ( , 0.6 m resolution) as well as of the Hexagon KH-9 panoramic camera ( , m resolution) was recently declassified in September 2011 by the US government, and provide with its vast data repository (that is however not available by now) further long term research opportunities also in regard of glaciological studies. ACKNOWLEDGEMENTS This study has been carried out within the framework of the research project WET (Variability and trends in water balance components on the Tibetan Plateau), funded by the German Federal Ministry of Education and Research (BMBF) within its program CAME (Central Asia Monsoon Dynamics and Geo-Ecosystem). Hexagon KH-9 and Landsat 7 ETM+ imagery was provided by the US Geological Survey, SRTM3 version 4 data was obtained from CGIAR (Consultative Group on International Agricultural Research), and ALOS- PRISM imagery was provided by JAXA (Japan Aerospace Exploration Agency) and purchased by GAF AG. TerraSAR-X data was provided by DLR (German Aerospace Center). We also acknowledge Tino

6 Pieczonka for his advice and support in data processing of KH-9 imagery. REFERENCES 1. Goldstein, R.M., Engelhardt, H., Kamb, B., Frolich, R.M. (1993). Satellite radar interferometry for monitoring ice sheet motion: Application to an Antarctic ice stream. Science, 262, Owen, L. A., Yi, C., Finkel, R. C. (2009). Quaternary glaciation of Muztag Ata and Kongur Shan: Evidence for glacier response to rapid climate changes throughout the Late Glacial and Holocene in westernmost Tibet. GSA Bulletin, 121(3/4), Pieczonka, T., Bolch, T., Wei, J., Liu, S. (2012). Heterogeneous mass loss of glaciers in the Aksu- Tarim Catchment (Central Tien Shan) revealed by 1976 KH-9 Hexagon and 2009 SPOT-5 stereo imagery. Remote Sensing of Environment (in revision). 4. Seong, Y. B., Owen, L. A., Yi, C., Finkel, R. C., Schoenbohm, L. (2009). Geomorphology of anomalously high glaciated mountains at the northwestern end of Tibet: Muztag Ata and Kongur Shan. Geomorphology, 103, Shangguan, D., Liu, S., Ding, Y., Ding, L., Libing, X., Cai, D.,, Zhang, Y. (2006). Monitoring the glacier changes in the Muztag Ata and Konggur mountains, east Pamirs, based on Chinese Glacier Inventory and recent satellite imagery. Annals of Glaciology, 43(1), Strozzi, T., Luckman, A., Murray, T., Wegmüller, U., and Werner, C. (2002). Glacier motion estimation using SAR offset-tracking procedures. IEEE Transactions Geoscience and Remote Sensing, 40(11), Surazakov, A. & Aizen, V. (2010). Positional accuracy evaluation of declassified Hexagon KH-9 mapping camera imagery. Photogrammetric Engineering & Remote Sensing, 76(5), Wang, Y., Ge, D., Hu, Q. and Guo, X. (2008). Surface subsidence monitoring using coherent point target SAR Interferometry. IEEE International Geoscience and Remote Sensing Symposium, vol. 4, Yan, S., Guo, H., Fu, W., Liu, G. & Ruan, Z. (2011). Kekesayi glacier velocity extraction based on the offsets derived from SAR images. In Geoscience and Remote Sensing Symposium (IGARSS) (pp ). Extended abstract. IEEE International. Vancouver, BC, Canada. 10. Yao, T., Thompson, L., Yang, W., Yu, W., Gao, Y., Guo, X.,, Joswiak, D. (2012). Different glacier status with atmospheric circulations in Tibetan Plateau and surroundings. Nature Climate Change, 2, Zebker, H. A. and Villasenor, J. (1992). Decorrelation in interferometric radar echoes., IEEE Transactions on Geosciences and Remote Sensing, 30(5),