Vol52,2,2009 167 ANNALS OF GEOPHYSICS, VOL. 52, N. 2, April 2009 Key words Synthetic Aperture Radar interferome- try – Independent Component Analysis – ground de- formations – landslides – subsidences 1. Introduction The ICA (Independent Component Analy- sis) technique is devised for separating signals obtained as a mixture of additive components with unknown mixing coefficients (e.g., Hyvärinen et al., 2001). More specifically, ICA algorithms are applied on a vector X of n exper- imental observations (x1, x2,..., xn), given by the overlapping of m different unknown signals (s1, s2,..., sm), or «sources», which are considered elements of a vector S, named the «vector of sources». The coefficients of the «mixing», aij, are considered the element of the matrix A, named «mixing matrix»: X = A·S (1.1) Given X, when the elements of S and of A are unknown, ICA algorithms are devised to ex- tract the best estimate of A and S, defined as a minus of an additional and multiplicative con- stant (Hyvärinen et al., 2001). In the case of SAR interferograms δφi, their phases for their n pixels (ordered in sequence of columns or rows) can be considered as one ob- servation xi: xi = δφi = (δφi1, δφi2,…, δφin) (1.2) In this way, an array of k interferograms can be considered as the columns of a matrix that can be treated as the vector of observation X: X = (x1, x2, ..., xk) = (δφ1, δφ2, …, δφk) (1.3) On the variations of InSAR-ICA altitudes in a mountain area of the Sele Valley (South Italy) Paola Ballatore Mediterranean Agency for Remote Sensing and Environmental Control (MARSec), Benevento, Italy Abstract The ERS-1/ERS-2 Synthetic Aperture Radar (SAR) data are processed by InSAR-ICA technique and the alti- tude variations are calculated for the period June-December 1995. For the same period and the same ERS- 1/ERS-2 SAR data, the nominal surface deformations are calculated by the standard DInSAR (Differential In- terferometric SAR) method, considered without any atmospheric correction. The results of the two different pro- cessing techniques are qualitatively compared with the previous geological and hydrological observations of the area under study. It is shown how the InSAR-ICA altitude variations are in better agreement with the expected characterization of this region. Although these variations cannot be considered quantitative precise surface de- formations at present, this finding is interpreted in terms of usefulness and reliability of InSAR-ICA altitude es- timations. Mailing address: Dr. Paola Ballatore, Mediterranean Agency for Remote Sensing and Environmental Control (MARSec), Villa dei Papi, Via Perlingieri 1, 82100 Bene- vento, Italy; e-mail: paolab@marsdb.it Vol52,2,2009 17-06-2009 19:02 Pagina 167 168 P. Ballatore Each δφi is given by the overlapping of phase shift components driven by different phe- nomena, such as the topographical altitude (δφtopo), the surface deformations (δφdef), the at- mospheric artifacts (δφatm) and generic noise (δφnoise): δφ i= δφtopo,i + δφdef,i + δφatm,i + δφnoise,i (1.4) Also other phenomena (e.g., orbital preci- sion, unwrapping, etc.) can drive minor phase shifts and can appear in δφi (Hanssen, 2001), but these can be neglected in the first instance or in specific observational circumstances. In eq. 1.4, each component can be considered due to a part common to the other interferograms (stopo, sdef, satm, etc), multiplied by coefficients, aij, depending on the specific interferogram. The common parts can be considered elements of the vector of sources (taken in sequence of columns or rows, similarly to X); the aij can be considered elements of the matrix A. X = (δφ1, δφ2, …, δφk) = = A·S = A·(stopo, sdef, satm, ...) (1.5) In this way, the identification of the differ- ent components of InSAR phase shifts can be successfully represented by the typical ICA problem solving, given in eq. 1.1 (Ballatore, 2006). In the case of SAR interferograms, the spe- cific FastICA algorithm (e.g., Hyvärinen and Oja, 1997) was previously employed in InSAR- ICA method (Ballatore 2006; 2007). After FastICA application on arrays of SAR interferograms, a different post-processing is necessary to pass from each source to the corre- sponding geophysical parameter. In particular, this post-processing has been devised for ex- traction of topographical altitudes, which re- sulted in good agreement with the standard SRTM-DTED (Shuttle Radar Topography Mis- sion – Digital Terrain Elevation Data) in specif- ic test sub-areas (Ballatore, 2006; 2007). For different SAR data sets, an extensive comparison between SRTM and InSAR-ICA altitudes has demonstrated that the distribution of the residuals is centered more or less close to zero (within the nominal errors), with distribu- tions whose standard deviations depend on the specific interferograms considered (Ballatore, 2009). In general, an agreement between In- SAR-ICA DEMs and SRTM altitudes is ob- served, after consideration of the criticality of SRTM measurements in the high-relief terrains (Falorni et al., 2005). This paper addresses the differences be- tween InSAR-ICA DEMs extracted from time- separated interferograms arrays. Here, these are not considered precise surface deformations (because not validated to date). However, the accuracy of this parameter as a more or less rough indication of surface deformations is qualitatively investigated in one particular case to highlighting elements on the correctness of InSAR-ICA DEMs and motivating any future works on processing their differences to extract information on terrain surface deformations. 2. The area under study The area considered here is a part of the Se- le valley and is illustrated in fig. 1, where the river flows near the SS91 at the central vertical location. This is a basin that is a tectonically active valley of South Italy, with typical slope instabilities and floods. The altitude changes are significant: the maximum elevations reach values >1400 m on the right and left sides and the minimum elevations reach about 150 m along the Sele river, passing through about 580 m a.s.l. at Senerchia and 590 m a.s.l. at Col- liano. The soil of this region is characterized by different lithotypes (Polemio and Petrucci, 2001; Wasowski et al., 2002). In particular, in- vestigations around the Sele river after the 1980 earthquake indicate higher concentration of landslides on the western than on the eastern slopes of the valley (Agnesi et al., 1983; Car- rara et al., 1986). In particular, hydrological studies also il- lustrate a substantial difference between the eastern and western sides of the valley: many more springs are located on the western side, with higher discharges, exceeding 50 l/s in a few cases (Budetta et al., 1988, and references Vol52,2,2009 17-06-2009 19:02 Pagina 168 169 On the variations of InSAR-ICA altitudes in a mountain area of the Sele Valley (South Italy) therein). Some major landslides were moni- tored and found active (with decrease in activ- ity) until 2001 (e.g., Polemio and Petrucci, 2001; and references therein). Because of the public availability of infor- mation on its peculiar terrain characterization, this area was chosen for qualitative comparison between InSAR DEM variations and the ground based geological/hydrological observa- tions. 3. SAR Data and Data Processing ERS-1 and ERS-2 Tandem data with track n. 265 and frame n. 2781 for the period from June to December 1995 were processed by In- SAR-ICA and DInSAR methods. The InSAR-ICA (Ballatore, 2007) was sep- arately applied on two arrays of interferograms, considering ERS-TANDEM couples of images acquired on: (A) June 2 and 3, July 7 and 8, Au- gust 11 and 12, 1995; (B) October 20 and 21, November 24 and 25, December 29 and 30, 1995. The differences between the DEMs ex- tracted from the interferograms array A and B were calculated and the interval between the maximum (positive value) and minimum (neg- ative value) differences divided into five sub- ranges (about equal among themselves) that are shown in different gray level in fig. 2 for about the same zone of Sele Valley shown in fig. 1. The black areas of fig. 2 correspond to the largest nominal subsidences, while the white areas correspond to the nominal uplifts. By considering the same ERS images spec- ified above and acquired on June 2 and 3 and on December 30, the standard three-pass DInSAR technique (e.g., Hanssen, 2001) was applied for estimating the nominal surface deformations during the same period June-December 1995. The DInSAR nominal surface deformations are shown in fig. 3 for the same area of fig. 2. Sim- ilar to fig. 2, fig. 3 shows a different gray scale level for each one of the five intervals of values between the maximum and the minimum phase differences, with black corresponding to maxi- mum nominal subsidences and white to the maximum nominal uplifts. The comparison between figs. 2 and 3 indi- cates that, according to InSAR-ICA DEMs differences, the maximum nominal subsi- dences are located mostly on the left of the Se- le river, whereas this is not so for the DInSAR results. It is worth mentioning that, in this process- ing, the DInSAR method does not include at- mospheric or spurious artifacts correction, which can be subsequently executed depending on user choice/methodology. In particular, by appropriate sorting of images with especially small atmospheric artifacts or specific choice of baselines or specific case by case processing, the three-pass DInSAR method is proven to give valuable and validated results. Specifically, in this case, the standard DInSAR nominal sur- face deformations without atmospheric correc- tions is intended to be compared with InSAR- ICA method, referring exactly to the same data availability and period. This is to show one of the uses of InSAR-ICA technique, that is its ca- pability of application in cases when no choic- es of data or ancillary different data availabili- ties are possible. Fig. 1. Area under study with indication of the Sele river location near the SS91. Vol52,2,2009 17-06-2009 19:02 Pagina 169 170 P. Ballatore 4. Observations The comparison between figs. 2 and 3 shows that in fig. 2 there is a more asymmetric distribution of the nominal maximum subsi- dences, which are located more on the western than on the eastern side of the river. Differently, fig. 3 shows a quite symmetric characterization on both sides of Sele. Therefore, the InSAR- ICA DEM variations are in better agreement with expectations from previous ground based investigations (see paragraph 2), which indicate larger landslides and springs occurrence on the west side of Sele than on the east side (Agnesi et al., 1983; Carrara et al., 1986; Budetta et al., 1988). This agreement is obtained for data and time period on which the corresponding DIn- SAR method would not give similar results. This is interpreted in terms of the capability of InSAR-ICA technique to provide useful DEM extraction in case of limited usability of atmos- pheric and ground data. Although at the present status, the InSAR- ICA DEMs-differences cannot represent pre- cise quantitative surface deformations, it can be of interest to note that the darkest colour areas in fig. 2 include alluvial deposits, characterized by middle permeability, with typical landslide prone areas (e.g., Polemio and Petrucci, 2001) and the most stable zones correspond to typical- ly stable high reliefs of limestone and dolomitic limestone (e.g., Wasowski et al., 2002). Finally, the specific highest positive increas- es in elevations of fig. 2 are located over vege- tated areas included in the Parco Regionale dei Monti Picentini on the west of Senerchia and the Riserva Naturale dei Monti Eremita- Marzano on the east of Valva, possible associ- ated with plants and trees growing during the summer season included between June and De- cember 1995. 5. Summary and Conclusions For an area of the Sele valley previous geo- logical/hydrological observations indicate larg- er landslides and springs on the West side than on the East side of the river (see paragraph 2). Fig. 2. Differences between InSAR-ICA DEMs re- lated to the period June-December 1995; the gray color ranges from maximum subsidence (black) to maximum uplifts (white). See text for more details. Fig. 3. Similar to fig. 2, but for the DInSAR nomi- nal surface deformations. 15.169° E 15.297° E 40.804° N 40.678° N 15.169° E 15.297° E 40.804° N 40.678° N Vol52,2,2009 18-06-2009 16:41 Pagina 170 171 On the variations of InSAR-ICA altitudes in a mountain area of the Sele Valley (South Italy) According to the possible ERS SAR data availability during the period June-December 1995, the DInSAR standard method (without atmospheric corrections) is applied and the nominal surface deformations are derived. Their distribution over the area of study is not in good agreement with expectations as the de- formations on the west are quite similar to those on the east of Sele. This indicates that over that specific period the DInSAR method cannot be routinely applied on the ERS SAR available da- ta alone. By using the same data available for the same period, two InSAR-ICA DEMs are de- rived for the time intervals June-August 1995 and September-December 1995. Their differ- ences have not been validated to represent pre- cise surface deformations, nevertheless they have to include these deformations, together with spurious artifacts. Unlike DInSAR observations, the distribu- tion of these InSAR-ICA DEM variations are in good agreement with previous geological/hy- drological observations over the area of study, showing larger subsidences taking place on the west of Sele. This finding is interpreted in terms of the good precision of InSAR-ICA DEM in this specific case, with errors evidently smaller than expected or about equal in the two esti- mates and eliminated by their subtractions. This case study confirms the usefulness of the InSAR-ICA method in circumstances of limited data availability and/or lack of ground based and meteorological ancillary data and supports the hypothesis of eventual possible ex- traction of surface deformations from differ- ences between time-spaced InSAR-ICA DEMs. Acknowledgements ERS-1 and ERS-2 data are provided by Eu- ropean Space Agency in the context of the Cat- egory-1 Program, project n. 3193. REFERENCES AGNESI, V., A. CARRARA, T. MACALUSO, S. MONTELEONE, G. PIPITONE and M. SORRISO VALVO (1983): Elementi tipologici e morfologici dei fenomeni di instabilità dei versanti indotti dal sisma del 1980 (alta valle del Sele), Geol. Appl. e Idrogeol., 18 (1), 309-341. BALLATORE, P. (2006): Synthetic Aperture Radar Interfer- ometry: separation of atmospheric artifacts from ef- fects due to the topography and the terrain displace- ments, Earth Planets and Space, 58, 927-935. BALLATORE, P. (2007): ICA technique applied to SAR inter- ferometry, Proceedings of the Envisat Symposium 2007, Montreux, Switzerland, April 23-27 2007, Euro- pean Space Agency, SP-636. BALLATORE, P. (2009): SAR interferometry: DEM estima- tions by Independent Component Analysis, to be sub- mitted. BUDETTA, P., P. CELICO, A. CORNIELLO, R. DE RISO, D. DUC- CI and P. NICOTERA (1988): Carta idrogeologica del F.186 (S. Angelo dei Lombardi) memoria illustrativa, Mem. Soc. Geol. It., 41, 1029-1038. CARRARA, A., V. AGNESI, T. MACALUSO, S. MONTELEONE and G. PIPITONE (1986): Slope movements induced by the southern Italy earthquake of November 1980, Geol. Appl. e Idrogeol., 21 (2), 237-250. FALORNI, G., V. TELES, E.R. VIVONI, R.L. BRAS and K.S. AMARATUNGA (2005): Analysis and characterizationof the vertical accuracy of digital elevation models from the Shuttle Radar Topography Mission, J. Geophys. Res., 110, F02005. HANSSEN, R. (2001): Radar Interferometry, (Kluwer Acad- emic Publishers, Dordrecht, The Netherlands). HYVÄRINEN, A. and E. OJA (1997): Fast fixed-point algo- rithm for independent component analysis, Neural Comp., 9, 1483. HYVÄRINEN, A., J. KARHUNEN and E. OJA (2001): Indepen- dent component analysis, (Wiley Interscience, New York, USA). POLEMIO, M. and O. PETRUCCI (2001): Hydrological moni- toring and image analysis of a mudslide in Southern Italy, Phys. Chem. Earth, 26, 689-695. WASOWSKY, J., V. DEL GAUDIO, P. PERRI and D. CAPOLON- GO (2002): Factors controlling seismic susceptibility of the Sele valley slopes: the case of the 1980 Irpinia earthquake re-examined, Survey of Geophysics, 23, 563-593. (received February 2, 2009; accepted March 25, 2009) Vol52,2,2009 17-06-2009 19:02 Pagina 171