Documenting Earthquake-Induced Liquefaction Using
Satellite Remote Sensing Image Transformations
Department of Geological Engineering, Michigan Technological University,1400 Townsend Drive, Houghton, MI 49931
LAURIE G. BAISE
Department of Civil and Environmental Engineering, Tufts University,Medford MA 02155
Alaska Satellite Facility, Geophysical Institute, University of AlaskaFairbanks,Fairbanks, AK 99775
Geophysical Institute, University of AlaskaFairbanks, Fairbanks, AK 99775
RAVI P. GUPTA
Department of Earth Sciences, Indian Institute of Technology,Roorkee 247667, India
Key Terms: Earthquake, Liquefaction, Remote Sens-ing, Bhuj
Documenting earthquake-induced liquefaction ef-fects is important to validate empirical liquefactionsusceptibility models and to enhance our understandingof the liquefaction process. Currently, after anearthquake, field-based mapping of liquefaction canbe sporadic and limited due to inaccessibility and lackof resources. Alternatively, researchers have usedchange detection with remotely sensed pre- and post-earthquake satellite images to map earthquake-inducedeffects. We hypothesize that as liquefaction occurs insaturated granular soils due to an increase in porepressure, liquefaction-induced surface changes shouldbe associated with increased moisture, and spectralbands/transformations that are sensitive to soil mois-ture can be used to identify these areas. We verify ourhypothesis using change detection with pre- and post-earthquake thermal and tasseled cap wetness imagesderived from available Landsat 7 Enhanced ThematicMapper Plus (ETM+) for the 2001 Bhuj earthquake inIndia. The tasseled cap wetness image is directly related
to the soil moisture content, whereas the thermal imageis inversely related to it. The change detection of thetasseled cap transform wetness image helped todelineate earthquake-induced liquefaction areas thatcorroborated well with previous studies. The extent ofliquefaction varied within and between geomorpholog-ical units, which we believe can be attributed todifferences in the soil moisture retention capacitywithin and between the geomorphological units.
Historically, liquefaction-related ground failureshave caused extensive structural and lifeline damagearound the world. Recent examples of these effectsinclude the damage produced during the 2001 Bhuj,India, 2010 Haiti, and 2010 and 2011 New Zealandearthquakes. It is observed from these earthquakesthat the occurrence of co-seismic liquefaction isrestricted to areas that contain saturated, near-surfacegranular sediments. Documenting earthquake-inducedliquefaction is important for developing case historydata sets of liquefaction. Recent work by Oommenet al. (2011) has shown that the bias due to thedifference in the ratio of the occurrence/non-occurrenceof liquefaction between the data set and population canadversely affect the ability to develop accurate proba-bilistic liquefaction susceptibility models (Oommen and
1Corresponding author: phone: 906-487-2045, fax: 906-487-3371,email: firstname.lastname@example.org.
Environmental & Engineering Geoscience, Vol. XIX, No. 4, November 2013, pp. 303318 303
Baise, 2010; Oommen et al., 2010). Thus, withimproved and more complete liquefaction case historydata sets, earthquake professionals will be able to refineexisting empirical prediction methods and enhancetheir understanding of liquefaction processes.
Currently, after an earthquake, the damagesinduced by the earthquake are documented byreconnaissance teams (Rathje and Adams, 2008).However, large earthquakes such as the 2001 M7.7Bhuj earthquake, which shook 70% of India, killed20,000 people, and made 600,000 people homeless,pose a great challenge to field reconnaissance teamsattempting to document the entire suite of liquefac-tion effects. These challenges are due to (1) inacces-sibility of sites immediately after the earthquake, (2)the short life span of surface manifestations ofliquefaction effects, (3) difficulty in mapping theareal extent of the failure, and (4) lack of resources.
Satellite remote sensing involves imaging thesurface of Earth in various spectral bands at differentspatial and temporal resolutions and can provide anunbiased record of events. Previous researchers haveused pre- and post-earthquake images to mapearthquake-induced effects (Gupta et al., 1995;Kohiyama and Yamazaki, 2005; Mansouri et al.,2005; Rathje et al., 2005; Huyck et al., 2006; Kayenet al., 2006; Rathje et al., 2006; and Eguchi et al.,2010). These images are often used to aid reconnais-sance teams (Hisada et al., 2005; Huyck et al., 2005).The current approach of computer-based processing ofthese images to support reconnaissance efforts fallsinto two general categories: (1) use of pre- and post-earthquake data to identify change, and (2) use of onlypost-earthquake imagery to identify damage. Thelimitation of the former approach is that if ideal pre-to post-earthquake image pairs are unavailable, thereis no quality measure of non-earthquake-inducedchanges (e.g., seasonal vegetation changes) for thefinal change detection map. The approach of thematicclassification using only post-earthquake images needssufficient training instances for the supervised classi-fication. The training instances of liquefaction-inducedsurface effects are only obtained from field reconnais-sance efforts. Therefore, neither approach sufficientlyaids the field reconnaissance teams in rapidly docu-menting liquefaction-induced surface effects.
The objective of this research is to describe andverify the utility of satellite remote sensing for aidingreconnaissance teams in better identifying liquefac-tion-induced surface effects using spectral bands/transformations that are sensitive to soil moisture.We hypothesize that because liquefaction occurs insaturated granular soils due to increase in porepressure, which often results in vertical flow of water,liquefaction-induced surface changes should have an
associated increase in soil moisture with respect tosimilar surrounding non-liquefied regions. The in-crease in soil moisture affects the signature in spectralbands that are sensitive to it, such as thermal infrared(TIR) and shortwave infrared (SWIR) (Yusuf et al.,2001). Additionally, components from special trans-forms, such as the tasseled cap transform wetnesscomponent, could potentially be suitable for identi-fying areas that have undergone increases in soilmoisture from earthquake-induced liquefaction. Thetasseled cap transformation is a useful tool to convertthe spectral reflectance to physical scene characteris-tics and was originally developed for understandingimportant phenomena of crop development in spec-tral space (Kauth and Thomas, 1976; Crist andCicone, 1984). In this study, we use the tasseled captransform component, which relates the spectralreflectance of Landsat data to surface wetness. Wetest our hypothesis using satellite imagery and withprevious studies from the Kutch region, in the state ofGujarat, western India, for the 2001 Bhuj earthquake(Figure 1).
On 26 January 2001, the Kutch region experiencedone of the most deadly earthquakes to strike India inits history. The epicenter of this intra-plate earth-quake, known as the Bhuj earthquake, was ,400 kmaway from the boundary of the Indian plate and,1,000 km away from the boundary of theHimalayan plate (Rastogi, 2004). The Kutch regionforms a crucial geodynamic part of the westerncontinental margin of the Indian subcontinent, and itis designated zone V (Figure 2) in the seismic zoningmap of India, having the highest earthquake risk.The Indian Standard (IS) building code assigns azone factor of 0.36 to the Kutch region, whichindicates that the buildings in that zone must bedesigned for forces associated with a horizontalground acceleration of 0.36g. It is evident fromFigure 2 that Kutch is the only region in Indiaoutside the plate-boundary region of the Himalayasthat is designated as zone V.
After the Bhuj earthquake, widespread appear-ances of water bodies, sand boils, and channels werereported by several reconnaissance teams and by localresidents in the Kutch region. The Indian remotesensing satellite (IRS-1C) with its Wide Field Sensor(WiFS) acquired an image 90 minutes after the 2001Bhuj earthquake (26 January, 8:46 a.m.). This imagecaptured near-real-time effects of the earthquake.Reconnaissance teams identified some of these asliquefaction and cited that the accessibility to severalregions of the affected area was poor, particularly
Oommen, Baise, Gens, Prakash, and Gupta
304 Environmental & Engineering Geoscience, Vol. XIX, No. 4, November 2013, pp. 303318
due to earthquake-induced damage (Bilham, 2001;Mohanty et al., 2001; Narula and Choubey, 2001;Saraf et al., 2002; Singh et al., 2002; and Ramakrish-nan et al., 2006). In this study, we use satellite images
acquired by the National Aeronautics and SpaceAdministration (NASA) Landsat Enhanced ThematicMapper to evaluate its use in mapping earthquake-induced liquefaction-related surface effects.
The study area (Kutch region) bears many geolog-ical similarities to the Mississippi Valley in the centralUnited States, which contains the New MadridSeismic Zone (Gomberg and Schweig, 2002). Lessonslearned from the Bhuj earthquake can be readilyapplied to other areas such as the Mississippi Valley.Large damaging earthquakes are rare in intra-platesettings like Kutch and New Madrid, and, hence,each intra-plate earthquake provides a unique oppor-tunity to make incremental advances in our ability toassess and understand the hazards posed by suchevents. Both of these regions have relatively flattopography, negating the need to correct the satellitedata for varying local an