Slow Deformation of Mt. Baekdu Stratovolcano Observed by Satellite Radar Interferometry

Slow Deformation of Mt. Baekdu Stratovolcano Observed by Satellite Radar Interferometry Sang-Wan Kim and Joong-Sun Won Department of Earth System Sciences, Yonsei University 134 Shinchon-dong, Seodaemun-gu, Seoul 120-749, Korea Fax) +82-2-2123-2673, E-mail: sangwan@ yonsei.ac.kr ABSTRACT Mt. Baekdu is a Cenozoic stratovolcano, where a series of micro-seismic events and gaseous emissions have been reported recently. Two-pass DInSAR technique was applied to detect possible displacement. Most interferometric phases out of 58 JERS-1 differential interferograms show concentric fringe patterns that correlate with elevation. From an analysis of fringe-duration relation, the fringe patterns can be interpreted mostly as an atmospheric contamination by stratified troposphere. To estimate the tropospheric delay, we used the data in Mt. Sobaek that is located about 20 km away from the summit of Mt. Baekdu and whose diameter is about 5 km. After removing the tropospheric effect, about 3 mm/yr of inflation was detected from 1992 to 1998. Although the volcanic inflation is not conclusive because of the large r.m.s. error, the results indicate that there exists a possibility of a slow upward deformation around the volcano. 1 INTRODUCTION Differential synthetic aperture radar interferometry (DInSAR) has proven to be a powerful technique for monitoring of subtle crustal deformation. The study on Mt. Etna s deformation by Massonnet et al. [1] that demonstrated for the first time the use of radar interferometry for volcano monitoring triggered many successive researches [2][3][4]. Atmospheric effects on mountainous area are known to be very significant. Accordingly, interferograms may reflect both deformation and tropospheric effects. For large-scale deformation fields, especially, tropospheric delay may produce volcano-wide effects up to several centimeters [3]. Mt. Baekdu is a historically active stratovolcano (60 km in diameter) located on the border between China and North Korea. A major eruption took place at 968±20 A.D., which was one of the largest eruptions ever since the human history [5]. Mt. Baekdu has been dormant since the last eruption in 1702. However, gaseous emissions, hot springs and minor earthquakes recently have been reported. Some Chinese scientists believe that the volcano is bulging extremely slowly Fig. 1. Left: location map of the study area. Right: SRTM-3 DEM. Boxes of red line and yellow line represent the areas covered by JERS-1 and ERS tracks. White rectangle inset is the study area. Triangle and dot denote the summit crater lake of Mt. Baekdu, and the summit of Mt. Sobaek. Proc. of FRINGE 2003 Workshop, Frascati, Italy, 1 5 December 2003 (ESA SP-550, June 2004) 57_kim

Fig. 2. JERS-1 SAR images set and interferograms. Dots and circles are the images with their acquisition dates and relative perpendicular baselines with respect of the first data. Solid dots represent the images used, and circles represent the images excluded from geodetic network analysis for troposphere. Thickness of lines reflects mean coherence of each interferogram. Left: 88/230 track, Right: 89/230 track. (about 5 mm per year). A summit caldera is 5-km-wide and 350-m-deep (the height difference is 850 m from the top of the mountain to the surface of lake), and is filled by a lake. Most surrounding areas are covered with vegetation and forest except the summit of volcano. In this study, we use ERS-1/2 C-band and JERS-1 L-band SAR data for monitoring of Mt. Baekdu, and discuss obtained fringe patterns. 2 DATA AND DInSAR PROCESSING ERS-1/2 (Track/Frame: 146/2763 and 375/2763) and JERS-1 (Path/Raw: 89-230 and 88-230) SAR data sets were used to detect surface deformation possibly occurred in Mt. Baekdu for a 10-year period (from 1992 to 2002). We have ten ERS-1/2 SAR and forty-one (track 88/230: 23 scenes, track 89/230: 18 scenes) JERS-1 SAR data sets. To apply twopass differential interferometry, a digital elevation model (DEM) was obtained by refining low resolution DEM using five interferograms of short span [6]. SRTM-3 DEM (~90-m spacing) has been released recently to the public, and its accuracy is about 6 m ( 1 σ ) [7]. We also used SRTM DEM to remove topographic contribution. A series of interferograms has been constructed by two-pass method. Fig. 2 shows a relative perpendicular baseline and acquisition date of JERS-1 SAR. Each line represents the interferogram. Most interferograms on winter season (during the period between late October and May) have very low coherences because the mountain is usually covered with snow during the period. Thus we have investigated mainly the images of summer season. In Fig. 2, solid dots represent the images used for the analysis, and circles represent the data sets excluded from network analysis for troposphere due to low coherence. Since the JERS-1 orbit is not accurate enough to estimate baseline from initial orbital information, ground control points that are generated from matching between simulated SAR and real SAR images are used for estimation the orbit. Finally we used Massonnet et al. [8] method for the JERS-1 SAR baseline fine-tuning process. The examples of topographically corrected interferograms are shown in Fig. 3 for ERS-1/2 and Fig. 4 for JERS-1. 3 ANALYSIS OF DIFFERENTIAL INTERFEROGRAM We obtained only two coherent pairs from ERS SAR data sets because of the temporal decorrelation due to forest and vegetation. A 70-day ERS interferogram shows a relatively clear fringe of a round shape. The differential phase that correlates with the topography of volcano may be caused by troposphere. Without any other observations such as GPS or meteorological data, a tropospheric effect cannot be removed by means of inspection of few interferograms. Therefore we use JERS-1 L-band SAR data sets for the study.

Fig. 3. ERS-1/2 Differential interferograms of (a) 9707/9709 pair (70-day), (b) 9508/9607 pair (351-day), and (c) 9806/0206 pair (1470-day). Color scale represents 2.8 cm displacement along the radar line-of-sight direction. Fig. 4. Differential Interferograms of JERS-1 SAR dataset corrected for topography around Mt. Baekdu volcano. Color scale represents one cycle of interferometric phase that can be interpreted as 11.8 cm displacement of the surface along the radar line-of-sight direction. Background image is the multi-image reflectivity map. Areas of loss of radar coherence are uncolored.

Fig. 5. Location map of the selected pixels for the modeling of 88_9209/9807 pair between elevation and phase, (a) at Mt. Baekdu, (b) at Mt Sobaek. Most interferometric phases out of 58 JERS-1 differential interferograms (some examples are shown in Fig. 4) show concentric fringe patterns correlated with the topography of volcano. The amount of fringe is not proportional to the perpendicular baseline and the duration of the SAR pairs, but it seems to correlate with their elevation even though at Mt Sobaek (refer to Fig. 5). Therefore most of the interferometric fringes could be regarded as tropospheric effect. The atmospheric heterogeneity is one of the potential error sources for DInSAR. A horizontally layered troposphere produces values of phase delay as a function of elevation [3][4], consequently concentric fringe patterns appear around Mt. Baekdu. To estimate the tropospheric delay, we aim to use the data in the surrounding areas. Mt. Sobaek is located about 20 km away from the summit of Mt. Baekdu, whose summit elevation and diameter is about 2,200 m and 5 km, respectively. Assuming that a deformation source is centered below the summit of Mt. Baekdu, its small diameter and the distance of 20 km can reduce the atmospheric signals. Mt. Sobaek is not always covered by the track of 89/230. It is possible to retrieve tropospheric information from the DInSAR image itself only in case of JERS-1 SAR data of the 88/230 track. 3.1 Fringe Counting We first select a set of coherent pixels from an averaged coherence map (Fig. 5). We then select a subset of pixels out of selected pixels that maintains the highest coherence over the duration of each interferogram. To distribute evenly in elevation and horizontal plane we divide study area into eight blocks around the summit of Baekdu, and then we select 5 most coherent pixels for every block in 50 meters elevation layer. For Mt. Sobaek 40 most coherent pixels are selected for every 50 meters layer. Fig. 5 is an example of 88_9209/9408 pair (track: 88/230, master: 1992/09/24, slave: 1994/08/29). The plots of differential phases with respect to elevations of SP (Fig. 6) show that the phases are strongly correlated with their heights. For fringe counting of each interferogram, we analyzed the correlation between pixel altitude and phase value. The following fitness function was used: i = j2 max R( m) e i i i π ( ϕ m ) obs ϕcal ( ) (1) ϕ (m) To define model phase cal, we use first-order polynomial because the height change from the base of the volcano edifice to the summit is relatively small (~1000 m). From the analysis of altitude-phase regression, we obtain polynomial coefficients of each interferogram. To obtain comparable quantity in two regions we define the phase computed from 1,400m to 2,600m as observed phase delay ( Pobs ) (refer to Fig. 6). The maximum Pobs reaches

Fig. 6. Examples of phase-elevation data and modeling. (a) 88_9209/9408 pair at Mt. Baekdu, (b) 88_9209/9408 pair at Mt. Sobaek. Solid dots are original data, and lines are the modeled phase delay. 7.39 radian(13.8 cm) with RMSE ±0.87 in 88_96089610 pair. The mean value Baekdu is 2.18 radian (4.1 cm) with RMSE ±0.93, and 3.2 Retrieval of Tropospheric Effect and Analysis Baekdu P obs of observed phase in Sobaek P obs is 2.01 radian (3.8 cm) with RMSE ±1.33. As shown in Fig. 2 there are closures among the different interferograms. Therefore it is possible to compensate each observed phase delay by means of the network adjustment. This network constitutes a typically overdetermined system with m observations and n unknowns: A ( m n) X( n 1) = Y( m 1) (2) The m observations are not fully independent. The rank of matrix A is n-1, consequently the only solution is unavailable from Eq. (2). We obtained the relative phase delays P cal in respect to the first acquisition data (92/09/24). This network adjustment allows to retrieve a relative phase delay for each single data from each interferograms and to decrease the observation error of individual DInSAR. Baekdu cal Fig. 7 shows the result of the network adjustment expressed as fringe delay for each image. The phase delay ( P ) Sobaek of each data estimated in Baekdu overlaps with the phase delay ( P cal ) of Sobaek in 1 σ. This demonstrates that most fringes observed in interferograms of Mt. Baekdu may be due to troposphere rather than actual deformation. Assuming that P cal Sobaek comprises only atmospheric effect, the difference between Pcal Baekdu Sobaek and P cal can be considered as the component of surface displacement. An abrupt motion on a large scale have not been reported, thus we expect a continuous slow deformation of Mt. Baekdu. The rate of the displacement is estimated using a straight line fit (refer to Fig. 7a). The phase shortening of 0.1 radian per year is computed in line-of-sight direction. If it is projected Sobaek in vertical movement the rate of inflation is about 2.4 mm/yr. The calculated differential delay P cal for each Baekdu interferograms can be used to subtract a tropospheric delay from observed value of P obs. The residual phase Baekdu Sobaek ( Pcal Pcal ) has a slight correlation with a time interval of interferograms (Fig. 7b). The gradient of a fitting line corresponds to the uplift of 2.6 mm/yr in the vertical direction. The correlation coefficients of two fitting lines in Fig. 7 are only 0.28 and 0.32. The confidence of the result is thus weak The possible maximum displacement in Mt. Baekdu can be more than 2.4 mm/yr (or 2.6 mm/yr) because it is an estimate from 1400 m to 2600 m. 4 CONCLUSIONS We apply the two-pass radar interferometry technique to ERS-1/2 and JERS-1 SAR data set for detecting possible slow surface deformation in Mt. Baekdu. ERS C-band SAR produces a very poor interferogram because of the temporal

Fig. 7. (a) Phase delay (from 1400 m to 2600 m) with error bar compensated for each image of JERS-1 88/230 pairs. Solid dots and red triangles correspond to phase delay estimated from Mt Baekdu and Mt. Sobaek, respectively. Solid line fits on the difference value (+ mark) between Mt. Baekdu and Mt. Sobaek, whose gradient corresponds to an uplift of 2.4 mm/yr. (b) Plot of time interval versus residual phase to subtract estimated phase of Sobaek from differential phase of Baekdu. The gradient of a fitting line indicates an uplift of 2.6 mm/yr. decorrelation due to forest and vegetation. However, JERS-1 SAR produces coherent interferograms even in pairs of long span (~ several years). The computation of 58 JERS-1 interferograms over Mt. Baekdu reveals a correlation between interferometric phases and topography that is related with tropospheric effect. The maximum and mean of the magnitudes of phase delay observed in Mt. Baekdu are 13.8 cm and 3.8 cm over 1200 m. Mt. Sobaek located in the surrounding regions of Baekdu volcano allows to estimate the tropospheric delay independent of volcano deformation. After removing the tropospheric effect, we obtained a displacement of about 3 mm/yr of inflation from 1992 to 1998. Although the large estimation error hinders determining the velocity of the deformation, the result indicates that a slow and upward moving deformation is in progress around the volcano. 5 REFERENCES [1] Massonnet D., Briole P., and Arnaud A., Deflation of Mount Etna monitored by spaceborne radar interferometry, Nature, Vol. 375, 567-570, 1995. [2] Amelung F., Jonsson S., Zebker H., and Segall P., Widespread uplift and trapdoor faulting on Galapagos volcanoes observed with radar interferometry, Nature, Vol. 407, 993-996, 2000. [3] Beauducel F., Briole P., and Froger J.-L., Volcano-wide fringes in ERS synthetic aperture radar interferograms of Etna (1992-1998): Deformation or tropospheric effect?, J. Geophys. Res., Vol. 105, 16,391-16,402, 2000. [4] Tarayre H. and Massonnet D., Atmospheric propagation heterogeneities revealed by ERS-1 interferometry, Geophys. Res. Lett., Vol. 23, 989-992, 1996. [5] Horn S. and Schmincke H.-U., Volatile emission during the eruption of Baitoushan volcano(china/north Korea) ca. 969 AD, Bulletin of Volcanology, Vol. 61, 537-555, 2000. [6] Seymour, M.S., Refining Low-Quality digital elevation models using synthetic aperture radar interferometry, The University of British Columbia, Ph.D. Dissertation, 1999. [7] Muller J.-P. and Backes D., Quality assessment of X- and C-SRTM with ERS-tandem DEMs over 4 European CEOS WGCV test sites, Frascati, Italy, 2003. [8] Massonnet D. and Feigl K. L., Radar Interferometry and its Application to Changes in the Earth s Surface, Reviews of Geophysics, Vol. 36, 441-500, 1998.