1999 Jiji earthquake

The 1999 Jiji earthquake, also known as the Chi-Chi earthquake or 921 earthquake, struck central Taiwan at approximately 1:47 a.m. local time on September 21, 1999. Its epicenter was in Nantou County, and rupture occurred principally along the Chelungpu thrust fault. Taiwan’s Central Weather Bureau reported a local magnitude of 7.3 and a focal depth of approximately 8 kilometres; published moment-magnitude estimates are approximately 7.6–7.7. The earthquake killed more than 2,400 people and caused widespread destruction of housing and infrastructure, particularly in central Taiwan. (scweb.cwa.gov.tw)

The disaster led to extensive rescue operations, resettlement and reconstruction, followed by changes to disaster-management institutions and seismic-design requirements. Its surface ruptures and strong-motion records also provided important evidence for research into fault mechanics, earthquake ground motion and structural performance. Scientific drilling subsequently investigated the fault’s physical properties, while preserved earthquake damage became the basis of the 921 Earthquake Museum of Taiwan. (nfa.gov.tw)

Regional tectonics and earthquake history

Taiwan occupies the collision zone between the Philippine Sea and Eurasian plates, where the Luzon volcanic arc meets the Eurasian continental margin. This collision connects two subduction systems of opposite polarity: south of Taiwan, South China Sea lithosphere descends eastward beneath the Philippine Sea Plate along the Manila Trench; to the northeast, the Philippine Sea Plate descends beneath the Eurasian margin along the Ryukyu Trench. GPS measurements reported in studies of the 1999 earthquake placed relative plate convergence at approximately 80–82 millimetres per year, directed northwestward. That motion is distributed across the island rather than accommodated on a single fault. Compression shortens and thickens the crust, producing Taiwan’s mountain ranges and its western fold-and-thrust belt, where sedimentary strata have been folded and displaced along predominantly east-dipping thrust faults. (agupubs.onlinelibrary.wiley.com)

The Chelungpu fault lies along the western foothills of central Taiwan, near the boundary between uplifted hills to the east and the Taichung Basin and adjoining alluvial lowlands to the west. Its principal trace trends approximately north–south, bending northeastward near its northern end. It is a west-verging thrust fault: the eastern hanging-wall block moves upward and westward relative to the western footwall. Seismological observations indicate an overall eastward dip of about $30^\circ$, although local geometry varies. Much of the fault follows bedding within the Pliocene Chinshui Shale. This geometry allows crustal shortening to be accommodated by older sedimentary rocks riding over younger deposits along the mountain front. (tem.tw)

Chelungpu belongs to a broader network of structures rather than forming an isolated plate boundary. The Changhua thrust lies farther west, while the Shuangtung fault lies to the east; the Taichung Basin occupies the intervening region between the Chelungpu and Changhua structures. In his 1999 geological compilation, Manuel G. Bonilla described Chelungpu as joining or becoming the Sani fault northward and connecting with the Tachienshan fault southward. These geological correlations must be distinguished from the extent of one earthquake’s rupture. Post-earthquake mapping documented more than 90 kilometres of surface rupture in 1999, generally following Chelungpu but also incorporating subsidiary structures, including a southern tear and a complex northern zone of folds and faults. A mapped geological fault, its inferred continuations and an individual earthquake’s surface breaks therefore need not have identical endpoints. (doi.org)

Western Taiwan had experienced several surface-rupturing earthquakes before 1999. Bonilla’s compilation recorded a rupture exceeding 13 kilometres during the March 17, 1906, Meishan earthquake, with substantial horizontal as well as vertical displacement, and a roughly 12-kilometre rupture during the December 5, 1946, Hsinhua earthquake. The Hsinchu–Taichung earthquake of April 21, 1935, produced two distinct rupture zones: predominantly horizontal movement along the southern Tunzijiao fault and predominantly reverse movement along the northern Shitan fault. The Central Weather Administration describes these zones as approximately 10 and 20 kilometres long respectively, with maximum vertical displacement reaching about 3 metres on Shitan. These events demonstrate that regional compression can generate both thrusting and strike-slip movement on differently oriented faults; they were not earlier repetitions of the entire 1999 Chelungpu rupture. (pubs.usgs.gov)

The historical comparison also depends on the measure of disaster severity. The Central Weather Administration records 3,276 deaths in 1935, exceeding the death toll of the 1999 earthquake, while identifying 1999 as the event causing the greatest physical and property damage. Consequently, descriptions of Jiji as Taiwan’s “most destructive” earthquake should not be read as claims that it was the deadliest. The 1935 earthquake provides both a tectonic precedent for surface faulting in west-central Taiwan and a distinct historical benchmark for mortality. (scweb.cwa.gov.tw)

Mainshock parameters and the earthquake sequence

The mainshock occurred at approximately 01:47 on September 21, 1999, in Taiwan standard time, equivalent to 17:47 UTC on September 20. Its epicenter was in Nantou County, near the town of Jiji, southwest of Sun Moon Lake. “Jiji” and “Chi-Chi” are alternative romanizations of the same place name, 集集; “921” refers to the earthquake’s local calendar date, September 21, rather than its UTC date. The event consequently appears under September 20 in international catalogs organized by UTC, but under September 21 in Taiwanese reports. (jsce.or.jp)

The Central Weather Bureau (CWB) reported a local magnitude of $M_L 7.3$ and a focal depth of 8.0 kilometres. Its archived felt-earthquake report gives an origin time of 01:47:15.9 and an epicentral location of $23.85^\circ\mathrm{N}, 120.82^\circ\mathrm{E}$. These values constitute a particular agency solution, not a universal set of parameters shared by every catalog. Published moment magnitudes include $M_w 7.6$, used in numerous seismological studies, and $M_w 7.7$, listed for the Harvard solution in the contemporary US Geological Survey monthly bulletin. That bulletin used a different origin time and hypocentral solution; its location and depth should therefore not be silently substituted into the CWB parameter set. (scweb.cwa.gov.tw)

The difference between 7.3 and 7.6–7.7 principally reflects different magnitude scales, whereas the difference between published moment magnitudes reflects separate source estimates. Local magnitude is calculated from recorded seismic-wave amplitudes, with corrections for distance and instrument response. Moment magnitude instead derives from seismic moment, a physical measure involving fault rigidity, rupture area and average slip. Amplitude-based scales have limited frequency and magnitude ranges; moment magnitude more reliably represents the size of very large earthquakes. Thus $M_L 7.3$ and $M_w 7.6$ are not successive measurements of an earthquake that increased in strength, nor should they be averaged into a single magnitude. (usgs.gov)

The aftershock sequence extended beyond the main rupture into surrounding structures. Relocation studies identified substantial activity east of the Chelungpu fault, including an east-dipping seismic zone near the Shuilikeng fault and a deeper, west-dipping zone beneath the western Central Range. Other aftershocks west of the middle Chelungpu segment had normal-faulting mechanisms, demonstrating that the sequence did not consist exclusively of renewed thrust movement on one fault plane. Kwang-Hee Kim and colleagues’ 2010 study described two $M_L 6.8$ aftershocks, approximately 1.5 and five days after the mainshock, and a damaging $M_L 6.7$ aftershock about a year later near the southern end of the fault. Aftershock activity generally declined over the following year and a half, although the Central Range showed a different temporal pattern. (sciencedirect.com)

Sequence statistics require explicit temporal and geographic limits. Kim and colleagues reported 13 significant aftershocks with $M_L \geq 6.0$ in the months following the earthquake. By contrast, the frequently cited total of 49,919 earthquakes refers to all earthquakes recorded throughout Taiwan during the calendar year 1999; 3,228 of these were classified as felt. It is therefore an annual regional seismicity total, not the number of Jiji aftershocks, and includes earthquakes outside the sequence as well as events preceding the mainshock. (sciencedirect.com)

Surface rupture and ground motion

The mapped surface rupture extended northward from the Zhushan area through central Taiwan toward Fengyuan, where its predominantly north–south trace turned east–west. Published lengths describe different mapped extents: Wen-Shan Chen and colleagues reported approximately 80 km along the Shihkang, Chelungpu and Tajienshan faults, Richard Heermance and colleagues described a 90-km rupture, and more extensive field studies counted over 100 km of breaks. These figures are not interchangeable measurements of a single straight line: coverage, the northern bend and the inclusion of subsidiary ruptures affect the total. The deformation included abrupt fault scarps, inclined ramps, folding and secondary fractures, rather than a continuous, simple vertical step. (sciencedirect.com)

Permanent displacement varied markedly along strike. In their 2003 comparison, Heermance, Zoe Shipton and James Evans reported average displacements of about 8 m in the northern region and 2 m in the south. Northern records contained less high-frequency ground motion, whereas southern shaking had a stronger high-frequency component despite smaller displacement. They proposed a relationship with fault structure: northern slip was concentrated within narrow clay gouge, while the southern fault occupied a broader zone of sheared shale. Surface measurements also require geometric care. A scarp’s height may represent only part of the vertical deformation where adjacent ground has folded; Chen and colleagues therefore measured structural relief beyond the folding zone. Neither a local scarp height nor a surface-displacement average is equivalent to slip inferred across the buried fault plane from seismic and geodetic models. (seismosoc.org)

The Central Weather Bureau’s strong-motion network captured an unusually dense record of the event. W. H. K. Lee and colleagues reported more than 60 recording sites within 20 km of the ruptures; their total exceeding 30,000 digital records included the mainshock and thousands of aftershocks, not the mainshock alone. Near-fault records showed pronounced velocity pulses, and motion was asymmetric across the fault: the eastern hanging wall—the block above the dipping fault—generally underwent greater displacement and larger peak velocities than the western footwall. Dynamic modelling by David Oglesby and Steven Day reproduced much of this contrast even with simplified stress conditions, demonstrating the importance of the dipping fault’s interaction with the free surface. Their models required additional spatial variation to reproduce the especially large northern motion. (pubs.usgs.gov)

The pulses constrain rupture timing and direction, but do not establish that every large northern pulse resulted from rupture directivity. Chen Ji and colleagues’ 2003 inversion combined velocity waveforms from 36 strong-motion stations with 119 GPS measurements and estimated an average rupture-front speed of approximately 2.0 km/s. Brad Aagaard, John Hall and Thomas Heaton subsequently showed that predominantly along-strike propagation on a thrust fault, as at Chi-Chi, limits strong shear-wave directivity because propagation is approximately perpendicular to the slip direction. Pulse interpretation therefore depends on fault geometry and slip direction, as well as the advancing rupture front. (pubs.usgs.gov)

The archived official intensity report places the strongest shaking in central Taiwan: Yuchi and Mingjian in Nantou are listed at intensity 7; Taichung city, Gukeng and Caoling in Yunlin, and Alishan in Chiayi at 6. Hualien city and several eastern stations reached 5, while Taipei and Magong in Penghu were listed at 4. These are local shaking intensities, not earthquake magnitudes or numerically equivalent Modified Mercalli ratings. The archive also requires a historical qualification: Taiwan’s scale originally ended at 6, covering peak ground accelerations above 250 gal. Intensity 7 was introduced on August 1, 2000, so the archived level-7 values reflect retrospective classification rather than the scale available when the earthquake occurred. (scweb.cwa.gov.tw)

Landslides, liquefaction and landscape change

Earthquake-triggered landslides ranged from shallow stripping of hillsides to deep rockslides that obstructed mountain valleys. At Tsaoling, also romanized as Caoling, approximately 125 million cubic metres of rock and soil moved across the Chingshui River. The failed slope consisted of Pliocene sandstone and shale, with bedding dipping downslope at about $14^\circ$. Field investigations by Masahiro Chigira and colleagues identified a smooth, nearly bedding-parallel sliding surface within thinly interbedded sandstone and shale or within shale beds. Their reconstruction distinguished the 1999 failure from simple reactivation of an older landslide: successive major failures at Tsaoling had developed progressively upslope, removing new portions of the mountain. (sciencedirect.com)

The Tsaoling slide crossed the valley floor and deposited material on the opposite slope, temporarily converting a river corridor into a large sediment barrier and impounded lake. Chigira and colleagues mapped a depositional area of approximately 3.4 square kilometres and described about 4 kilometres of buried river channel. A later study by Rou-Fei Chen and colleagues characterized the obstruction as a roughly 5-kilometre-long dam; these dimensions describe the deposit’s extent along the valley, not its height. Blocks exceeding 10 metres in diameter occurred within the debris. Subsequent river incision cut into the obstruction, so the landslide’s effects included both immediate valley blockage and continuing downstream release of sediment as the deposit eroded. (sciencedirect.com)

On lowland alluvial deposits, soil liquefaction produced a different class of ground failure. Cyclic shaking increased pore-water pressure in susceptible saturated sediments, reducing the effective stress between grains and therefore their resistance to deformation. Investigations documented sediment boils, foundation-bearing failures, settlement and lateral spreading in Nantou, Wufeng and Yuanlin, as well as liquefaction in coastal hydraulic fills. Settlement involved downward ground or foundation movement, whereas lateral spreading displaced ground horizontally toward riverbanks or other unsupported margins. These processes were consequences of shaking and local soil conditions, rather than direct displacement across the Chelungpu fault. (sciencedirect.com)

Wufeng illustrates the importance of shallow groundwater and riverbank geometry. Its alluvial plain is crossed by several streams, and groundwater generally lay only 1–2 metres below the surface. Reconnaissance recorded building settlement and tilting in developed areas, while lateral spreading along Dry Creek River and Dove Nest Creek was commonly accompanied by sediment boils. At the investigated Wufeng Site C, an open rice paddy subsided and spread toward an adjacent creek; the water table was approximately 1 metre deep, and investigators identified potentially susceptible sediments between depths of about 1 and 7 metres. The spatial separation between these failures and mapped fault rupture was important for interpreting cracks and offsets correctly. (escholarship.org)

Landscape disturbance continued through subsequent rainy seasons. Loose landslide debris supplied channels with sediment, while earthquake-damaged slopes remained susceptible to rainfall-triggered failure. In the Chenyoulan River catchment, mapped landslide area increased from approximately 12.9 square kilometres after the earthquake to 24.4 square kilometres following Typhoon Toraji in 2001 and 31.6 square kilometres following Typhoon Mindulle in 2004. These were successive mapped extents, not areas all displaced during the mainshock. Niels Hovius and colleagues’ 2011 analysis of landslide maps and suspended-sediment records found that enhanced mass wasting and river sediment evacuation peaked at more than five times background rates, then progressively approached pre-earthquake levels over about six years. The sedimentary response therefore depended on the interaction between earthquake-produced material and later storms, rather than shaking alone. (sciencedirect.com)

Casualties, displacement and economic losses

Casualty figures changed substantially during the emergency and differ from later retrospective totals. The International Federation of Red Cross and Red Crescent Societies (IFRC), in its situation report of October 4, 1999, recorded 2,192 deaths, 8,735 people described as seriously injured, and 97 still missing or trapped in collapsed buildings. A study by S. N. Han published in the Taiwan Epidemiology Bulletin in 2000 cited National Fire Administration statistics as of October 11, 1999: 2,329 deaths and 8,722 injuries. In contrast, the Central Weather Administration’s retrospective earthquake FAQ lists 2,456 deaths and 10,718 injuries, while a January 2020 National Museum of Natural Science conference paper gives 2,415 deaths and more than 11,000 injuries. These are distinct reporting snapshots, not interchangeable components of a single total; in particular, early missing-or-trapped counts cannot simply be added to later death tolls. (ifrc.org)

Losses were concentrated in Nantou and the then Taichung County, but the locations with the largest numbers of deaths were not necessarily those with the highest population-relative mortality. Han’s township-level study found 181 deaths in Puli, the largest toll within Nantou County, whereas Zhongliao’s mortality approached 1% of its population. In Taichung County, Dongshi—rendered “Tungshi” in the study—recorded 322 deaths, while Shigang had a mortality rate of 1.14%, the highest among the townships examined. These rates used deaths directly attributed to the earthquake and year-end 1998 populations. Fatal building collapses also occurred farther away: the IFRC’s October 4 report recorded more than 80 deaths at Taipei’s Tunghsing building, with twelve people still believed buried there. (cdc.gov.tw)

Displacement estimates measured a different consequence from structural damage. The IFRC estimated approximately 100,000 people homeless on October 4, but explicitly noted that no official registration had yet established how many shelter occupants had actually lost their homes. Some residents stayed outside because of aftershock fears or because electricity and water supplies were unavailable despite their houses remaining intact. Thus, displaced people, homeless households, destroyed dwellings and damaged housing units were separate statistical categories. An October 12 Reuters assessment preserved by the Asian Disaster Reduction Center reported about 51,000 buildings toppled or damaged but 82,238 individual households with uninhabitable accommodation. The Central Weather Administration’s retrospective FAQ instead lists 53,661 houses fully destroyed and 53,024 damaged; those figures describe its housing-damage categories, not numbers of displaced residents. (ifrc.org)

Contemporary monetary assessments likewise combined different kinds of loss. Reuters’ October 12 report presented an accounting dated October 11 comprising NT$203.21 billion in collapsed or damaged buildings, NT$11.45 billion in electricity, gas, petroleum and water facilities, and NT$15.79 billion in roads, railways, ports and telecommunications. It separately listed NT$61.56 billion in lost industrial revenues associated with damaged production systems or power interruptions. Although the report grouped these under “direct losses,” industrial revenue forgone is conceptually different from the physical destruction of buildings and infrastructure: it measures interrupted economic activity rather than damaged assets. (adrc.asia)

The National Museum of Natural Science’s retrospective account gives approximately NT$300 billion in financial losses. A broader estimate in the Earthquake Engineering Research Institute’s 2001 reconnaissance report, edited by Christopher Arnold, placed damage including lost productivity at US$20–30 billion. The latter range explicitly extends beyond physical damage, and therefore cannot be treated as a currency conversion of the NT$300 billion figure or as an additional loss to be added to it. (libknowledge.nmns.edu.tw)

Buildings and lifeline infrastructure

Damage to reinforced-concrete housing frequently concentrated in open ground floors beneath more heavily partitioned residential storeys. Street-front buildings combined shops or parking with housing above, leaving comparatively few walls at street level. These soft-storey configurations concentrated lateral deformation in ground-floor columns, sometimes causing the upper floors to drop or tilt together. The California Seismic Safety Commission’s June 2000 reconnaissance findings identified soft first storeys as a recurrent collapse mechanism, while noting that buildings complying with construction standards generally performed well. Column deficiencies documented in damaged buildings included widely spaced transverse reinforcement and ties terminating in 90-degree hooks, which provided inadequate confinement as concrete deteriorated under repeated loading. (ssc.ca.gov)

Schools exhibited a related but distinct short-column effect. Masonry or concrete infill beneath classroom windows restrained the lower portions of columns, leaving only a short exposed length to accommodate lateral movement. The resulting high shear demand could produce brittle column failure before substantial flexural deformation developed. Reconnaissance described this damage particularly along the classroom façades parallel to exterior corridors. Repeated use of similar classroom layouts reproduced the same vulnerability across numerous schools; the MAA engineering reconnaissance reported approximately 786 schools damaged, a category encompassing different degrees of damage rather than complete destruction alone. (maa.com.tw)

Bridge damage reflected several mechanisms that should not be conflated. Shaking and abutment movement could impose shear forces on piers and dislodge supported spans; at the Toong-Tour Bridge, investigators proposed that movement of the northern abutment toward the river, combined with longitudinal shaking, contributed to pier failures. Near the Dajia River, direct fault crossing imposed permanent offsets: reconnaissance photographs of a bridge west of Shihkang Dam recorded approximately 3–4 metres of lateral displacement and 4–5 metres of uplift. East of the dam, two spans of an eleven-span bridge collapsed and piers tilted approximately one degree. On Highway 3 between Fengyuan and Dongshi, a separate collapse occurred where a landslide pushed ground near the northern abutment toward the creek. These observations distinguish structural response to vibration from displacement of the ground supporting the bridge. (eeri.org)

Shihkang Dam suffered localized structural failure where the rupture crossed its foundation, rather than uniform collapse of the entire dam. Engineering reconnaissance reported approximately 9.8 metres of uplift on its eastern portion against 2.1 metres on the western portion, producing about 7.7 metres of differential movement. The displaced sections could no longer retain water normally. The California commission reported that the dam supplied about 40 percent of Taichung County’s water, making the fault-crossing failure a major disruption to regional water infrastructure. (maa.com.tw)

The electrical grid transmitted disruption beyond the principal damage zone. Damage at the Chungliao switchyard and to transmission towers interrupted northward power transfers; the New Zealand reconnaissance reported rationing affecting 3.6 million households, with daily outages of up to six-and-a-half hours. Its account placed island-wide restoration on October 10, 1999, although temporary connections remained heavily loaded. Industrial interruptions extended to Hsinchu’s technology firms, well away from the rupture: the MAA report recorded restoration of full power to the industrial park on September 25, after outages had interrupted semiconductor and silicon-wafer production. (bulletin.nzsee.org.nz)

Rail transport also suffered damage to lines and associated infrastructure. In telecommunications, the New Zealand team reported service disruption to 198,700 subscribers—about eight percent of central Taiwan’s subscribers—with heavy damage to external network plant and power-supply equipment. Communications failures therefore involved both physical network damage and dependence on electricity. Research on emergency communications subsequently identified base-station damage, power loss and broken backhaul connections among the mechanisms disabling mobile communications during the earthquake, constraining the networks available for disaster response. (ssc.ca.gov)

Rescue and immediate humanitarian response

Residents and survivors began extracting people from collapsed buildings before organized rescue teams arrived. Local firefighters and volunteer groups followed during the first hours of September 21, 1999, while county and municipal authorities organized relief operations. According to President Lee Teng-hui’s October 5 account, the Ministry of National Defense ordered military rescue teams to mobilize at 02:30, with initial units reaching affected areas around 04:00. Deployment expanded during the day, adding military personnel to civilian operations. On September 22, a supervisory center convened by Vice President Lien Chan was established to coordinate central and local agencies. (eng.buffalo.edu)

Access and information failures constrained this mobilization. Reconnaissance observers documented how collapsed bridges and landslides obstructed transport, while power failures disrupted telecommunications needed to assess damage and dispatch assistance. The difficulty was not simply reaching individual collapsed buildings: damage was dispersed across numerous communities. (eng.buffalo.edu) Coordination also remained uneven after teams arrived. Taiwan’s National Fire Agency later acknowledged shortcomings in handling international contacts, equipment and operational arrangements. A 2019 retrospective in the Journal of Acute Medicine described rescue teams arriving without adequate coordination and relief supplies whose quantity did not compensate for poor matching to local needs. (ebook.nfa.gov.tw)

Overseas assistance supplied specialist urban search-and-rescue personnel, dogs and equipment for searching collapsed structures. The National Fire Agency’s 2020 retrospective recorded 38 international teams from 20 countries, comprising 728 personnel and 103 dogs. (ebook.nfa.gov.tw) Operational records show how particular contingents were employed. Singapore’s rescuers established their base at Fengyuan City Fire Station and initially divided between Fengyuan and Dali before concentrating operations in Dali on September 22. The Singapore Civil Defence Force’s account records the rescue of an eight-year-old boy from a collapsed 12-storey building, followed by the joint extrication, with Taiwanese firefighters, of a man trapped in a five-storey golf clubhouse. (scdf.gov.sg)

United States assistance was organized through the United States Agency for International Development following a Taiwanese request transmitted through the American Institute in Taiwan. USAID’s September 22 deployment announcement distinguished a 19-person advance group, including Miami-Dade rescue personnel and disaster-assistance specialists, from a subsequent 73-person group incorporating Fairfax County rescuers. The latter departed Dover Air Force Base aboard a C-5 aircraft carrying 106,000 pounds of equipment, three vehicles and four search dogs. The advance group’s assignment included arranging reception, technical support and logistics for the incoming contingent; the announced US$2.3 million covered transportation of the teams and emergency equipment. (web-archive-2017.ait.org.tw)

Immediate accommodation consisted of collective shelters and dispersed tent camps rather than completed replacement housing. Local governments and the army opened shelters in schools and other public buildings; volunteers prepared meals, while public agencies and voluntary organizations supplied sanitation, medical care and other services. Informal camps also occupied parks, parking areas and spaces beside homes. These emergency arrangements were distinct from subsequent resettlement options: prefabricated temporary housing, assistance to rent existing dwellings, and access to public housing. A contemporary reconnaissance report reproduced the Executive Yuan’s September 28 relief schedule, which offered temporary accommodation or a rental subsidy of NT$3,000 per person per month, while also describing favorable public-housing purchase terms for eligible owners of destroyed or damaged homes. (eng.buffalo.edu)

Housing, community and regional reconstruction

President Lee Teng-hui’s emergency decree of September 25, 1999, allowed reconstruction authorities to simplify land-use, building and procurement procedures and authorized exceptional financing. A longer-term statutory framework followed with the promulgation of the Temporary Act for Post-921 Earthquake Reconstruction on February 3, 2000. The act assigned coordination, review, decision-making and supervision to an Executive Yuan reconstruction commission, while county and municipal governments organized local implementation, including reconstruction plans and land-use approvals. Central ministries supplied sectoral programmes and funding; civil organizations contributed donations, professional assistance and construction projects rather than operating through a single government delivery system. (english.president.gov.tw)

Resettlement offered several routes rather than requiring every displaced household to enter a prefabricated settlement: temporary housing, rental subsidies and access to public housing. These arrangements supported families while homes were repaired, strengthened, rebuilt or replaced through purchase. Government housing measures included preferential loans, credit guarantees and relief on existing mortgages. An Executive Yuan report in 2005 recorded 32,702 approved applications, worth approximately NT$57.8 billion, under the NT$100 billion emergency home-reconstruction mortgage programme. Nationwide, 5,854 prefabricated units were constructed at 112 sites; these were transitional accommodation, not the total number of replacement homes required. (nmns.edu.tw)

Rebuilding apartment complexes was particularly difficult because individual ownership of units intersected with shared land and common structures. Owners had to agree on reconstruction, redistribute property rights, negotiate with contractors and secure financing despite differing incomes and outstanding debts. Dispersal into temporary accommodation also complicated collective decisions. Pei-Chun Shao’s 2002 study identified residents’ associations and professional support as important mechanisms for maintaining contact and negotiating reconstruction; it distinguished rebuilding on the original site at the original scale from reconstruction through urban-renewal procedures. In April 2001, the government directed the 921 Earthquake Reconstruction Foundation to use private donations for the Linmen programme, providing interest-free financing to overcome funding bottlenecks. A 2007 presidential account reported assistance to 63 apartment communities containing 5,174 households. (jstage.jst.go.jp)

School and community recovery combined public works with projects undertaken by volunteers and nongovernmental organizations. Tzu Chi’s Project Hope repaired or rebuilt 51 elementary and junior-high schools, while its volunteers also participated in transitional housing and continuing support for affected communities. Government programmes extended beyond buildings to livelihoods: the September 21 Earthquake Work Relief Programme began in 1999, followed by the Disaster Area Reconstruction Force Employment Programme in 2000. Agricultural recovery included restoring irrigation and helping producers market local products; an Executive Yuan report recorded 87 agricultural-product promotional events in 2003 and 52 in 2004. These activities accompanied the restoration of schools, community facilities and local services. (fundmerger.tzuchi.org.sg)

The approaching end of the original reconstruction period did not mean that all households or landscapes had recovered. On September 22, 2004, the Executive Yuan explicitly identified remaining temporary-housing residents, apartment reconstruction and comprehensive river rehabilitation as tasks requiring continued work by the responsible ministries. The temporary act was subsequently extended to February 4, 2006. A 2005 government report still counted 808 households living in prefabricated accommodation and described preferential rehousing arrangements for low- and lower-middle-income residents. Regional rehabilitation also addressed earthquake damage compounded by Typhoons Toraji and Nari: integrated plans covered 7,176 square kilometres across the Zhuoshui, Dajia, Da’an and Wu river basins, coordinating upstream soil conservation and forestry with downstream water-management works. (ey.gov.tw)

Disaster-management and engineering changes

The earthquake accelerated the consolidation of Taiwan’s disaster-management system, although administrative reform had begun before 1999: the National Fire Agency (NFA) was established in 1995. The Disaster Prevention and Protection Act, promulgated on July 19, 2000, provided a statutory framework covering prevention, emergency response and post-disaster recovery, rather than rescue alone. It assigned responsibilities across central, municipal or county, and township authorities. On August 25, 2000, the Executive Yuan established its Disaster Prevention and Protection Commission to coordinate ministries, with administrative support from the NFA. These arrangements linked coordination between government levels with coordination among agencies responsible for different hazards and services. (mag.nfa.gov.tw)

Communications reform addressed both inter-agency connectivity and congestion caused by public inquiries. An NFA account reported that calls to police and fire emergency centres during the earthquake reached 11–27 times normal volumes, with requests to locate relatives competing with urgent rescue calls. Subsequent arrangements included dedicated satellite and microwave communications systems linking the three administrative levels, and mobile command vehicles capable of integrating communications equipment at disaster sites. The separate 1991 emergency-message service provided a channel for reporting personal safety. Mutual-support arrangements also enabled neighbouring jurisdictions to supply personnel and equipment when an affected county or city could not cope with its own resources. These were institutional provisions for later emergencies, distinct from the improvised coordination of September 1999. (nfa.gov.tw)

Specialized rescue capability developed through permanent organization and training. Following legislation expanding the NFA’s organization in 2001, its Special Search and Rescue Team began establishing operational units in 2002; the eastern unit at Taitung’s Fengnian Airport was inaugurated on June 25. This created a standing capability separate from the domestic and overseas teams mobilized immediately after the earthquake. Professional training subsequently gained a dedicated national base at Zhushan, Nantou County, where the NFA Training Center formally opened on January 19, 2010. Its first-phase facilities included 66 simulation sites in 13 categories, supporting practical instruction in building fires, road accidents, water rescue, mountain search and rescue, and hazardous-material incidents. (nfa.gov.tw)

Engineering changes combined stronger requirements for new construction with measures for existing buildings. In December 1999, the construction authorities revised seismic-design provisions using newly available strong-motion data. The 2005 code subsequently refined seismic zoning to township-level areas, incorporated near-fault effects and introduced maximum-considered-earthquake spectra alongside design-earthquake spectra. These changes made design demands more sensitive to location and earthquake characteristics rather than relying solely on broad regional classifications. For schools, researchers Keh-Chyuan Tsai and Shyh-Jiann Hwang described a Ministry of Education–adopted assessment sequence comprising an administrator’s survey, professional preliminary evaluation and detailed engineering analysis. Experimentally verified retrofitting techniques included reinforced-concrete or steel column jackets and additional wing walls beside columns; review of engineering designs and inspection during construction formed part of the procedure. (scweb.cwa.gov.tw)

The continued vulnerability of older buildings required programmes extending well beyond reconstruction in the original disaster area. In May 2019, the Executive Yuan described a nationwide 2019–2021 assessment and strengthening programme with a planned budget of approximately NT$5.4 billion. Its rapid-screening component targeted about 36,000 reinforced-concrete buildings of six or more storeys constructed before December 31, 1999, using permit drawings to identify potentially weak storeys. Detailed assessment could lead to strengthening, replacement or urban renewal. A staged-retrofit subsidy covered 45% of eligible strengthening costs, capped at NT$2.2 million per project, to reduce collapse risk while comprehensive redevelopment was being organized. The programme explicitly distinguished screening from engineering assessment and strengthening from replacement; post-1999 design improvements did not themselves upgrade the older building stock. (ey.gov.tw)

Seismological research and scientific drilling

Researchers reconstructed the earthquake’s rupture by combining strong-motion waveforms, permanent ground displacements measured by the Global Positioning System (GPS), and field mapping of the surface break. These observations supplied complementary constraints: waveforms recorded the timing of motion, GPS measurements constrained the final deformation, and mapped ruptures helped define fault geometry. In a 2003 study, Chen Ji, Donald Helmberger, David Wald and Kuo-Fong Ma jointly inverted three-component velocity records from 36 strong-motion stations and 119 GPS measurements using a three-plane fault model. They estimated slip amplitude, direction, initiation time and duration on individual fault patches, then tested the reconstruction against independent teleseismic records. The results showed that bends in the fault influenced rupture development and that peak slip velocities were generally higher near the surface than at depth. These were model-derived histories, rather than direct measurements of underground motion. (pubs.usgs.gov)

The Taiwan Chelungpu-fault Drilling Project (TCDP) investigated the physical conditions associated with the northern rupture’s unusually large displacement. Drilling at Dakeng, near Taichung, took place during 2004–2005. Hole A reached approximately 2,003 metres, while the neighbouring Hole B, about 40 metres away, reached approximately 1,350 metres. Hole A provided continuous cores below approximately 432 metres, allowing researchers to examine the fault rocks directly rather than infer their properties solely from surface observations. A fault zone at approximately 1,111 metres depth, designated FZA1111, was identified as the zone that accommodated the 1999 slip. (gep.ncu.edu.tw)

Core descriptions were integrated with downhole measurements of compressional- and shear-wave velocity, electrical resistivity, density, natural gamma radiation, temperature and borehole-wall imagery. Yun-Hao Wu and colleagues’ 2008 analysis used these records to distinguish ten geophysical subunits and relate changes in seismic anisotropy to bedding, fractures and shear zones. This comparison helped separate variations caused by rock type from those associated with deformation. The study also suggested that FZA1111 could have acted as a fluid conduit after the earthquake; this interpretation concerned post-earthquake conditions, not a direct observation of fluid flow during the mainshock. (scholars.ncu.edu.tw)

Borehole temperatures supplied an independent constraint on frictional resistance. Yasuyuki Kano and colleagues reported in 2006 a temperature anomaly of approximately $0.06\,^\circ\mathrm{C}$, spread across roughly 40 metres and centred on the fault near 1,100 metres depth, six years after the earthquake. Interpreting this residual signal as heat generated by the 1999 slip implied low frictional resistance during movement. Geochemical work published in 2008 identified sharp variations in fluid-mobile elements and strontium-isotope ratios within intensely sheared bands 2–15 centimetres thick. The authors interpreted these patterns as evidence of interaction between fault material and fluids heated above approximately $350\,^\circ\mathrm{C}$. That inferred transient temperature was distinct from the small residual anomaly subsequently measured in the borehole. (agupubs.onlinelibrary.wiley.com)

Laboratory friction and permeability measurements were used to test thermal pressurization as a weakening mechanism: frictional heating expands pore fluids, raising fluid pressure and reducing the effective normal stress that resists sliding. Wataru Tanikawa and Toshihiko Shimamoto’s 2009 experiments and numerical modelling suggested that differences in fluid-transport properties could make this process more effective in the northern fault segment than in the south. Such results supported a proposed explanation for the contrasting rupture behaviour; they did not constitute a measurement of pore-pressure rise during the earthquake itself. (agupubs.onlinelibrary.wiley.com)

Paleoseismic trenching extended the record beyond instrumental observations. Wen-Shan Chen and colleagues excavated across the middle Chelungpu fault, including a pineapple field, and documented evidence for at least four earthquakes over approximately 1,900 years, including 1999. Radiocarbon dates and historical evidence bracketed the preceding event between approximately 430 and 150 years before the study. Their reconstruction yielded an average recurrence interval shorter than 700 years and a long-term slip rate of about 8.7 millimetres per year. These estimates depended on identifying separate deformation episodes and measuring displacement beyond the distributed folding zone. A minimum event count and broad age brackets constrain past activity, but do not establish a regular cycle or a date for a future rupture. (pubs.usgs.gov)

Preserved sites, commemoration and education

The former Guangfu Junior High School campus in Wufeng, now part of Taichung, preserves earthquake damage where it occurred rather than reconstructing it elsewhere. The Chelungpu fault rupture crossed approximately 340 metres of the campus, passing through the athletic field, school buildings and entrance. The displaced running track and collapsed classrooms became central exhibits of the 921 Earthquake Museum of Taiwan, administered by the National Museum of Natural Science. The combination of fault displacement, damaged structures and an uplifted riverbed helped distinguish this location from other proposed memorial sites. The museum officially opened on September 21, 2004, the earthquake’s fifth anniversary. (nmns.edu.tw)

Preservation required intervention rather than simply leaving the ruins untouched. Public ownership of the school grounds facilitated retention of the site, while the damaged classroom buildings were stabilized using methods intended to retain their visible post-earthquake appearance. Protective membrane canopies were installed over the outdoor running track and the damaged northern classroom block. These shelters themselves required maintenance: in 2022, work included membrane replacement, steel-cable tension checks and renewed anticorrosion coatings. The track’s polyurethane surface helped retain a legible record of its deformation despite exposure to wind and rain. These measures preserve both the earthquake’s physical traces and the conditions needed for public access to them. (nmns.edu.tw)

The museum’s five principal exhibition areas connect geological processes with building performance, personal experience and recovery: the Chelungpu Fault Gallery, Earthquake Engineering Hall, Image Gallery, Disaster Prevention Hall and Reconstruction Records Hall. The fault gallery combines the preserved rupture with explanatory and interactive displays. The engineering hall, planned with participation from the National Center for Research on Earthquake Engineering, uses working models to explain seismic isolation and energy-dissipation devices. Images and reconstruction records extend the interpretation beyond physical damage to the disaster’s human consequences and subsequent rebuilding. The campus also acquired a practical training function: in early 2018, the museum began cooperating with the Taichung fire service to conduct rescue-dog training in its collapsed northern classrooms. (nmns.edu.tw)

September 21 is observed in Taiwan as National Disaster Preparedness Day, linking remembrance of those killed in 1999 with recurring public exercises. School drills translate that commemoration into rehearsed actions. On September 21, 2020, for example, primary and secondary schools conducted synchronized exercises at 9:21 a.m., following simulated earthquake information transmitted through the Central Weather Bureau’s earthquake early-warning system. Participants practised “Drop, Cover, and Hold On.” The Ministry of Education also documented exercises at a school for students with hearing impairments, including assistance and guidance from teachers, illustrating how preparedness activities were adapted to different pupils’ needs. (english.ey.gov.tw)

Exercises associated with the anniversary need not occur on September 21 itself. The 2024 national exercise was scheduled for September 20 at 9:21 a.m., using a simulated magnitude-7.3 earthquake in Zhuqi Township, Chiayi County, and resulting disasters across western Taiwan. Such scenarios give the annual observance an operational dimension beyond memorial ceremonies, allowing public warning and disaster-response arrangements to be exercised against a specified hypothetical event. (nfa.gov.tw)

References

1. Surface Rupture of 1999 Chi-Chi Earthquake Yields Insights on Active Tectonics of Central Taiwanseismosoc.org
2. Geometry and structure of northern surface ruptures of the 1999 Mw = 7.6 Chi-Chi Taiwan earthquake: Influence from inherited fold belt structuresresearchoutput.ncku.edu.tw
3. USGS Open-File Report 99-447pubs.usgs.gov
4. U.S. Geological Survey Open-File Report 99-447pubs.usgs.gov
5. FAQ for Earthquake—What has been the most catastrophic earthquake disaster in Taiwan?scweb.cwa.gov.tw
6. 地震百問—臺灣最嚴重的震災情形如何？scweb.cwa.gov.tw
7. 3 Main Shock and Aftershocksjsce.or.jp
8. 中央氣象署數位科普網-震源、震央，傻傻分不清！pweb.cwa.gov.tw
9. Chichiarchive.bukohome.com
10. CWB free-field strong-motion data from the 21 September Chi-Chi, Taiwan, earthquakeusgs.gov
11. 中央氣象局有感地震報告scweb.cwa.gov.tw
12. Preliminary Determination of Epicenters: September 1999pubs.usgs.gov
13. Magnitude Typesusgs.gov
14. Moment magnitude, Richter scale - what are the different magnitude scales, and why are there so many?usgs.gov
15. Intelligent Earthquake Alarm System with Self-calibration Mechanismwww-en.ntut.edu.tw
16. Paleoseismology of the Chelungpu Fault during the past 1900 yearsusgs.gov
17. Fault Structure Control on Fault Slip and Ground Motion during the 1999 Rupture of the Chelungpu Fault, Taiwanseismosoc.org
18. CWB free-field strong-motion data from the 21 September Chi-Chi, Taiwan, earthquakepubs.usgs.gov
19. Slip history and dynamic implications of the 1999 Chi-Chi, Taiwan, earthquakepubs.usgs.gov
20. Effects of fault dip and slip rake angles on near-source ground motions: Why rupture directivity was minimal in the 1999 Chi-Chi, Taiwan, earthquakeusgs.gov
21. 20周年專刊-各觀測系統發展史scweb.cwa.gov.tw
22. Central Weather Bureau research report MOTC-CWB-109-E-06scweb.cwa.gov.tw
23. Large earthquake-triggered landslides and mountain belt erosion: The Tsaoling case, Taiwancomptes-rendus.academie-sciences.fr
24. Documentation of soil conditions at liquefaction and non-liquefaction sites from 1999 Chi-Chi (Taiwan) earthquakeescholarship.org
25. Prolonged seismically induced erosion and the mass balance of a large earthquakenora.nerc.ac.uk
26. Mortality of the 921 Earthquake in Nantou and Taichung Countiescdc.gov.tw
27. Chi-Chi, Taiwan, Earthquake of September 21, 1999: Reconnaissance Reportbooks.google.com
28. Seismic Safety Commission Findings: Lessons Learned from Recent Earthquakes in Turkey, Greece, and Taiwanssc.ca.gov
29. Shaking Table Tests of RC Frames with Non-ductile Detailing of Columns and Jointsscholars.lib.ntu.edu.tw
30. The Chi-Chi, Taiwan Earthquake of September 21, 1999: EERI Special Earthquake Reporteeri.org
31. Taiwan Earthquake—Bridges across Tachia River and to the South along Highway 3geerassociation.org
32. The Chi-Chi, Taiwan Earthquake of 21 September 1999: Report of the NZSEE Reconnaissance Teambulletin.nzsee.org.nz
33. Challenges of Emergency Communication Networkcs.nccu.edu.tw
34. The 1999 Taiwan Earthquakeeng.buffalo.edu
35. President Lee's Opening Remarks Made at the Press Conferenceenglish.president.gov.tw
36. Examining the Disaster Response Twenty Years After the 1999 Chi-Chi Earthquakepmc.ncbi.nlm.nih.gov
37. The U.S. Agency for International Development (USAID) Activates Search and Rescue Operations for Earthquake Relief in Taiwanweb-archive-2017.ait.org.tw
38. President Lee decrees emergency ruleenglish.president.gov.tw
39. 制定九二一震災重建暫行條例president.gov.tw
40. 九二一震災重建暫行條例-全國法規資料庫law.moj.gov.tw
41. 921 Earthquake Museum of Taiwan-National Museum of Natural Sciencenmns.edu.tw
42. 行政院施政報告ey.gov.tw
43. Post-quake Condominium Reconstruction for Taiwan Chi-Chi Earthquake in 1999jstage.jst.go.jp
44. 總統接見「921地震63處集合式住宅都市更新會理事長及總幹事」president.gov.tw
45. Introduction - Tzu-Chi Foundation (Singapore)fundmerger.tzuchi.org.sg
46. Multiple Employment Development Programmol.gov.tw
47. 行政院第2907次院會決議ey.gov.tw
48. 九二一震災重建暫行條例uract.nlma.gov.tw
49. 消防署走過30年 打造強韌消防 精彩呈現各項成果(上)mag.nfa.gov.tw
50. FAQ for Earthquake—19. What is the classification of earthquake-resistant standards for buildings in Taiwan?scweb.cwa.gov.tw
51. Seimic Retrofit Program for Taiwan School Buildings After 1999 Chi-Chi Earthquakeiitk.ac.in
52. 全國建築物耐震安檢及重建補強—給國人一個安心的家ey.gov.tw
53. Coseismic Slip Distribution of the 1999 Chi-Chi, Taiwan, Earthquake Deduced from Inversion Analysis of GPS Dataseismosoc.org
54. Core-log integration studies in hole-A of Taiwan Chelungpu-fault Drilling Projectgep.ncu.edu.tw
55. Stratigraphy and sedimentology of the Taiwan Chelungpu-fault Drilling Project (TCDP) boreholesbasin.earth.ncu.edu.tw
56. Coseismic fluid–rock interactions at high temperatures in the Chelungpu faultnature.com
57. Paleoseismology of the Chelungpu Fault during the past 1900 yearspubs.usgs.gov
58. 簡介—921地震教育園區nmns.edu.tw
59. 戶外跑道保存介紹—921地震教育園區nmns.edu.tw
60. Annual nationwide earthquake drills to strengthen disaster resilienceenglish.ey.gov.tw