At 2:46pm on 11 March 2011, a Mw 9.0 megathrust earthquake ruptured 500km of the Japan Trench off the Pacific coast of Tohoku. The seafloor displaced vertically by several metres across an area the size of Connecticut. The resulting tsunami reached heights of 40 metres in the channelling bays of Iwate Prefecture, overtopped seawall defences designed for much smaller events, and penetrated up to 10km inland across the low-gradient Sendai plain. Nearly 16,000 people died. The Fukushima Daiichi nuclear plant was inundated, producing a secondary catastrophe that continued for years.
Total economic losses from the TΕhoku earthquake and tsunami exceeded USD 210 billion β the largest economic loss from any natural disaster in recorded history at that time. Insured losses were approximately USD 35β40 billion β a protection gap of over 80%. Of the insured loss, a substantial fraction came from the earthquake ground shaking component, not the tsunami. Japan's commercial earthquake insurance market had broad coverage; Japan's commercial tsunami insurance was largely embedded in government schemes and poorly penetrated at the commercial level.
Tsunami is a rare but catastrophic peril that operates on a fundamentally different timescale from most other natural hazards. The same earthquake that triggers ground shaking lasting seconds can initiate a tsunami that arrives at a distant coastline hours later β giving populations warning time that earthquake provides to no one. Yet the insured loss from tsunami can equal or exceed the ground shaking component for coastal portfolios, while remaining almost entirely invisible in standard earthquake cat model output.
A tsunami is a series of long-wavelength ocean waves generated by the sudden displacement of a large volume of water. Unlike wind-driven surface waves, which involve only the upper layer of the ocean, tsunami waves involve the entire water column from seafloor to surface. This gives them enormously greater energy per unit wavelength than surface waves of comparable height β a tsunami in deep ocean may be less than a metre high but travel at 700β900 km/h, carrying energy equivalent to thousands of surface waves.
Tsunamis can be generated by three primary mechanisms:
Vertical seafloor displacement from a large subduction zone earthquake displaces the overlying water column. The magnitude threshold for significant tsunami generation is approximately Mw 7.5; megathrust events above Mw 8.5 generate the most destructive tsunamis. Strike-slip earthquakes (horizontal motion) generate little tsunami even at high magnitude β vertical displacement is the critical parameter.
Rapid mass failure of seafloor sediment or coastal cliff material displaces water. Can generate locally devastating tsunamis even without an associated earthquake. The 1958 Lituya Bay megatsplash (triggered by a rockfall into a fjord) reached 524m β the highest recorded tsunami run-up in history. Submarine landslides on continental margins can generate far-field tsunamis, though typically smaller than major earthquake-generated events.
Caldera collapse, flank failure, or explosive eruption of a shallow-marine volcano can displace large water volumes. The 2022 Hunga TongaβHunga Ha'apai eruption generated a tsunami that reached coastlines across the Pacific within hours. Pyroclastic flows entering the ocean also generate localised tsunamis. Volcanic tsunamis are less predictable than earthquake-generated ones and may not trigger standard seismic warning systems.
Tsunami propagation follows two distinct phases with very different physical characteristics. In deep ocean, tsunami waves travel at speeds governed by water depth β approximately 700β900 km/h in the deep Pacific β but with wave heights of less than a metre. The waves are essentially invisible to ships and undetectable without specialised sensors. As the wave enters shallower water approaching the coast, the wave train slows dramatically due to reduced water depth. The wave's energy β conserved from the deep ocean β is compressed into a shallower, slower wave: heights increase to metres or tens of metres while wavelength compresses from hundreds of kilometres to tens of kilometres. This shoaling process is the primary amplification mechanism.
Coastal geometry then determines local run-up height. Open, gently sloping coasts dissipate energy over a broad front. V-shaped bays and fjords funnel and concentrate the wave, producing run-up heights that may be 3β5Γ the open-coast value from the same event. Resonance effects in enclosed bays can cause successive waves to reinforce each other, making later waves in a tsunami train more destructive than the first. The lesson from every major tsunami event is that local coastal geometry determines which communities experience catastrophic inundation and which experience only moderate flooding β from the same source event.
Near-field event β warning time severely limited. 2011 TΕhoku: first wave arrived before many evacuation routes were cleared. Seawall defences overtopped before populations fully evacuated.
Near-field for outer coast communities. Puget Sound populations have more warning time but tsunami energy is partially dissipated by the Strait of Juan de Fuca. A Cascadia megathrust remains the US Pacific Northwest's most significant uninsured catastrophe risk.
Far-field event β significant warning time available from Pacific Tsunami Warning Center. 1960 Chilean earthquake generated a far-field tsunami that killed 61 people in Hawaii and 122 in Japan, despite many hours of warning.
2004 Indian Ocean tsunami: near-field for Sumatra (minutes); far-field for Sri Lanka, India, and Thailand (2β3 hours). Warning systems were absent in the Indian Ocean in 2004. IOTWS (Indian Ocean Tsunami Warning System) established afterwards.
Tsunami hazard modelling is the most computationally intensive of any natural peril cat model. It requires simulating the full physical sequence from earthquake source rupture through deep-ocean propagation to near-shore shoaling and coastal inundation β at a resolution fine enough to capture the coastal geometry effects that determine run-up height at specific locations. The calculation chain involves physics at three very different spatial scales: the fault rupture (hundreds of kilometres), the ocean propagation (thousands of kilometres), and the coastal inundation (tens to hundreds of metres).
Tsunami hazard begins with characterising the tsunami source β the submarine earthquake, landslide, or volcanic event that generates the wave. For earthquake-generated tsunamis, this requires not just the magnitude and location of the event but the full rupture geometry: the length, width, and slip distribution across the fault plane, and the resulting pattern of vertical seafloor deformation. Two earthquakes of identical magnitude can produce very different tsunamis if the slip is concentrated in the shallow portion of the fault (generating more vertical seafloor displacement) versus distributed across a wider, deeper rupture plane.
The stochastic source catalogue for a tsunami model samples the full distribution of earthquake rupture scenarios on each subduction zone, producing tens of thousands of synthetic events with varying magnitudes, rupture locations, and slip distributions. Each synthetic event produces a different initial sea surface perturbation, which then propagates through the ocean model. The catalogue must adequately represent the tail of the magnitude distribution β including rare Mw 9.0+ megathrust events β because these events dominate the 1-in-500 year and 1-in-1000 year tsunami loss estimates.
The propagation model routes the initial sea surface disturbance through the ocean using the shallow water wave equations β a system of partial differential equations that describe how wave energy travels as a function of water depth. In the deep ocean, propagation models are accurate and computationally efficient. The challenge is the transition to shallow water and the coastal inundation zone, where the model must resolve individual bays, headlands, and river channels at sub-100m resolution to capture the local amplification effects that determine run-up height at specific locations.
A tsunami model that resolves the coastline at 1km grid spacing will produce smoothed run-up estimates that miss the 3β5Γ amplification in V-shaped bays and fjords and the 50% reduction on exposed headlands. For a portfolio spread along a 200km coastline, a 1km-resolution model may assign broadly similar inundation depths across the entire coast, while the actual event produces catastrophic inundation in two bays and negligible inundation on the headlands between them. For accurate per-location loss estimation, coastal inundation models require resolution of 50β100m or better β but running a 50m resolution inundation model across an entire coastline for tens of thousands of synthetic events is computationally prohibitive. Most commercial tsunami models use a nested grid approach: coarse resolution in the deep ocean, progressively finer near the coast, with full high-resolution inundation computed only for pre-identified high-risk coastal segments.
The historical instrumental record of tsunamis is far too short to reliably estimate return periods for large events at specific coastlines. The instrumental record extends perhaps 150 years for the most data-rich basins (Pacific) and much less for others (Indian Ocean). But the return period of a Mw 9.0+ megathrust event on a specific subduction zone may be 300β1,000 years β meaning the instrumental record contains at most one or two such events, providing almost no statistical basis for frequency estimation. Palaeotsunamis β the geological record of past tsunami inundation preserved in coastal sediment sequences β extend the record by thousands of years but at coarser temporal resolution and with significant uncertainty in run-up height estimation.
Tsunami damage to structures operates through mechanisms fundamentally different from those of any other natural peril. Where wind, earthquake, and wildfire damage primarily through pressure, acceleration, and heat respectively, tsunami damages through hydrodynamic force, hydrostatic pressure, debris impact, and scour β all simultaneously, in water that may be laden with vehicles, boats, building debris, and sediment.
Unlike flood, where inundation depth is the primary intensity metric and velocity is secondary, tsunami vulnerability requires both depth and flow velocity as primary inputs. Deep, slow-moving inundation β typical of river flood β primarily damages buildings through water ingress and contents loss. Shallow but fast-moving tsunami inundation β which can reach 5β10 m/s in channelling terrain β generates hydrodynamic drag forces that can overturn and displace structural elements, move vehicles and boats into buildings, and strip foundations through scour. A building that would survive a 1.5m deep river flood can be destroyed by 0.5m of fast-moving tsunami inundation if the flow velocity is sufficient to generate debris impact loads.
Indicative for well-engineered RC frame construction. Timber, masonry, and light steel frame structures fail at significantly lower inundation depths. Flow velocity amplifies damage at all depth levels β 3m depth at 5 m/s flow velocity is more destructive than 5m depth at 0.5 m/s. Based on post-2011 TΕhoku field survey data.
| Variable | High Vulnerability | Low Vulnerability | Mechanism |
|---|---|---|---|
| Construction class | HIGH Timber frame, unreinforced masonry, light steel | LOW Reinforced concrete frame, shear wall | Hydrodynamic drag and debris impact β wood and masonry fail at flow velocities that RC frames resist |
| Foundation type | HIGH Shallow spread footings, slab-on-grade in sandy soil | LOW Deep pile foundations into bedrock or dense soil | Scour removes soil from around shallow foundations; pile foundations resist scour and uplift forces |
| Building orientation | HIGH Long face perpendicular to flow direction | LOW Narrow face to flow, or open ground floor (piloti) | Drag force is proportional to frontal area exposed to flow. Open ground floors allow water to pass through rather than imposing full hydrodynamic load on the structure |
| Number of storeys | HIGH Single-storey β full inundation likely at moderate wave heights | LOW Multi-storey β upper floors above inundation level provide vertical evacuation refuge and reduce total loss | Multi-storey buildings limit the proportion of TIV exposed to inundation; also serve as vertical evacuation points during the event |
| Coastal position | HIGH Within 500m of coastline, low elevation, V-shaped bay | LOW Above maximum run-up elevation, behind effective seawall for sub-threshold events | Run-up height and flow velocity both decrease rapidly with elevation and distance from coast β the most important determinant of whether a structure is in the inundation zone at all |
Tsunami has the most fragmented insurance coverage of any major natural peril β and in most markets outside Japan and New Zealand, it is largely absent from the commercial insurance landscape entirely. The combination of very low frequency, very high severity, and extreme geographic concentration makes tsunami an extremely difficult peril to price and pool in private markets without government involvement.
Japan operates a two-layer residential earthquake and tsunami insurance system. The Japanese Earthquake Reinsurance (JER) scheme provides government-backed residential earthquake insurance that includes tsunami as a covered peril β the private market provides the first layer, the government backstop absorbs losses beyond specified thresholds. Following the 2011 TΕhoku event, residential earthquake/tsunami policyholders received JER payments across affected prefectures. Commercial earthquake and tsunami insurance operates separately through the private commercial market, where coverage availability and pricing vary by insurer and account. The 2011 event produced a significant commercial tsunami insured loss, but the residential JER system absorbed the largest single component of the residential loss.
New Zealand's Earthquake Commission (EQC) provides first-layer residential earthquake and tsunami coverage up to NZD 300,000 per dwelling, with the private market sitting above the EQC cap. New Zealand faces tsunami risk from both near-field sources (the Hikurangi Subduction Zone off the North Island's east coast, capable of Mw 8.5+ events) and far-field sources (South American and Pacific subduction zones). The near-field Hikurangi tsunami scenario β with warning times of 10β30 minutes for some East Coast communities β represents a scenario where EQC tsunami coverage would be triggered simultaneously with earthquake coverage, producing a compound loss that stresses the system's total capacity.
Outside Japan, New Zealand, and a handful of other markets, commercial tsunami insurance is either unavailable, extremely limited, or embedded in earthquake policies without explicit acknowledgement. In the Indian Ocean rim β where the 2004 tsunami demonstrated catastrophic exposure across Sri Lanka, India, Thailand, and Indonesia β insurance penetration for tsunami was negligible before the event and has improved only modestly since. In the Pacific Northwest US β which faces the Cascadia Subduction Zone scenario β tsunami coverage for coastal commercial properties is not widely available or purchased, and standard commercial property policies typically exclude flood (the mechanism under which tsunami damage would be categorised).
In most markets outside Japan, tsunami damage to a commercial property would be classified as a flood loss under standard policy language β and flood is commonly excluded from standard commercial property all-perils policies or subject to severe sub-limits. A coastal industrial portfolio in the Pacific Northwest US, the Philippines, or Indonesia that suffers tsunami inundation may find that its standard commercial property policy excludes the loss as a flood event, even if earthquake coverage is present. The earthquake that triggered the tsunami may be a covered peril; the tsunami damage may not be. This coverage gap is a significant risk for Lloyd's and reinsurance underwriters writing coastal commercial property in tsunami-exposed markets without explicit tsunami coverage verification.
| Region | Source Zones | Insurance Significance | Modelling Maturity |
|---|---|---|---|
| Japan | Japan Trench (Pacific Plate subduction) β Mw 9+ megathrust capable. Nankai Trough β repeat Mw 8β8.5 events on ~100β150 year cycle; next major event considered overdue. Multiple crustal fault systems with local tsunami potential. | The most significant tsunami insurance market globally. JER residential system provides broad coverage. Commercial tsunami exposure is large β Japan has enormous coastal industrial and commercial infrastructure. TΕhoku 2011 remains the benchmark commercial tsunami loss event. | Very high. Best-developed tsunami hazard models in the world. Extensive post-2011 field survey data. Nested high-resolution inundation models for most major coastal cities. PTHA (Probabilistic Tsunami Hazard Analysis) methodology most advanced for Japan. |
| Pacific Northwest (US/Canada) | Cascadia Subduction Zone β 1,300km fault capable of Mw 9.0+ megathrust events. Return period estimated at 200β500 years; last major event in January 1700. Near-field threat to Oregon and Washington coasts; far-field threat to California and British Columbia. | Very high uninsured exposure β large coastal commercial and industrial value with minimal tsunami insurance coverage. Portland and Seattle have significant insured property but most commercial policies exclude flood. A Cascadia megathrust is considered one of the most significant uninsured catastrophe scenarios in North America. | Good for hazard; limited for insured loss. NOAA and USGS have produced detailed Cascadia tsunami inundation maps. Commercial tsunami cat models available from specialist vendors. Insured loss modelling is constrained by the near-total absence of tsunami insurance in the commercial market. |
| Indian Ocean Rim (Indonesia, India, Sri Lanka, Thailand) |
Sunda Trench β the source of the 2004 Boxing Day tsunami (Mw 9.1). Active subduction along Sumatra and Java coasts. Multiple Mw 8.5+ events possible across the Sunda arc. The Indian Ocean now has IOTWS warning infrastructure established post-2004. | Low insurance penetration despite catastrophic demonstrated exposure. 2004 event caused USD 10B+ economic loss with minimal insured loss. Tourism and coastal resort infrastructure in Thailand has improved coverage since 2004. Broader coastal commercial and residential market remains largely uninsured. | Moderate. IOTWS provides warning infrastructure but commercial tsunami cat models for Indian Ocean are less well-developed than Pacific. Palaeotsunamis provide recurrence data for the Sunda arc but at coarser resolution than Japan. Vulnerability calibration limited by absence of systematic post-event loss surveys. |
| New Zealand | Hikurangi Subduction Zone (North Island east coast) β capable of Mw 8.5+ events with short warning times for coastal communities. Alpine Fault (South Island) β primarily strike-slip, limited direct tsunami potential but triggers submarine landslides. Far-field Pacific sources. | EQC provides first-layer residential coverage including tsunami. Commercial market above EQC cap. Significant coastal commercial and tourism infrastructure in Bay of Plenty and Hawke's Bay regions. Hikurangi near-field scenario with 10β30 minute warning times is the critical design scenario for emergency management and insurance modelling. | Good and improving. GNS Science has produced detailed Hikurangi PTHA. Post-Canterbury and KaikΕura events have calibrated earthquake model components. Dedicated Hikurangi tsunami inundation models available at high resolution for key coastal towns. |
| Mediterranean | Eastern Mediterranean subduction zones (Hellenic Arc, Cyprus Arc) capable of Mw 7.5β8.0 events. Historical record includes the 365 CE Crete earthquake and tsunami that devastated coastal communities across the Eastern Mediterranean. Limited far-field exposure from Atlantic sources. | Low but growing awareness. Coastal tourism and commercial infrastructure in Greece, Turkey, and the Levant has significant exposure. Insurance penetration for tsunami is minimal. A major Eastern Mediterranean event affecting resort and commercial coastlines would cause large economic losses with very limited insured loss recovery. | Low to moderate. Historical and palaeotsunamis provide recurrence data but instrumental record is short. Commercial tsunami cat models for the Mediterranean are limited. The growing coastal development in the region has outpaced model development and insurance market awareness. |
The instrumental record of tsunamis extends approximately 150 years in the best-documented basins β far shorter than the return periods of major megathrust events (200β1,000 years). Estimating the 1-in-500 year tsunami loss from a 150-year record requires extrapolation across four times the observation window. Palaeotsunamis extend the record but introduce uncertainty in event timing, magnitude, and run-up height. For any tsunami-exposed portfolio, the epistemic uncertainty in the tail of the loss distribution is substantially larger than for earthquake, where the historical record at least includes representative samples of the most damaging magnitude classes.
Accurate per-location tsunami loss estimation requires high-resolution coastal inundation modelling (50β100m grid) that captures bay amplification and run-up variation along the coast. Running such models for tens of thousands of synthetic events across an entire subduction zone is computationally prohibitive. Most commercial tsunami models use pre-computed inundation grids for a set of scenario events, supplemented by interpolation for intermediate scenarios. This approach introduces approximation errors that are largest for locations where coastal geometry produces strong local amplification β exactly the locations where tsunami hazard is highest and per-location accuracy most important.
The vast majority of commercial tsunami cat models are calibrated to earthquake-generated tsunamis on major subduction zones. Submarine landslide-generated tsunamis β which can be locally catastrophic without any associated earthquake large enough to trigger standard warning systems β are not modelled in most commercial platforms. Volcanic tsunami, demonstrated at scale by the 2022 Tonga event, is also absent from standard commercial model releases. For portfolios in regions where landslide or volcanic tsunamis represent a non-trivial hazard β Norwegian fjords, Caribbean volcanic islands, Pacific island chains β the earthquake-only tsunami model is capturing only a portion of the total tsunami risk.
In most markets outside Japan and New Zealand, the question of whether tsunami damage is covered by a commercial property policy depends on policy language that was typically drafted without explicit reference to tsunami. If flood is excluded and tsunami damage is classified as flood, the physical loss is not an insured loss β producing a modelling scenario where the cat model estimates a large loss but the gross financial module loss is near zero due to the coverage gap. Models that do not explicitly account for policy flood exclusions will overstate the insured loss from tsunami events in markets where this exclusion is common.
The most consequential operational limitation: most commercial earthquake cat model runs do not include tsunami by default. An analyst who runs the Japan earthquake model and presents the output as "the Japan earthquake loss" is producing a number that excludes tsunami β which for coastal portfolios can be comparable to or larger than the ground shaking component for the most extreme events. The 2011 TΕhoku insured loss included both components but they were modelled and reported separately by most vendors. For any coastal portfolio in a tsunami-exposed region, the practice of running earthquake and tsunami models separately and presenting the combined view is essential β but is often not standard practice for time-pressured underwriting workflows.
A series of long-wavelength ocean waves generated by sudden displacement of a large volume of water β primarily by submarine earthquake, but also by submarine landslide and volcanic activity. Involves the full water column, not just the surface layer. Travels at 700β900 km/h in deep ocean; slows and amplifies as it approaches the coast.
A subduction zone earthquake occurring on the interface between two tectonic plates β the largest category of earthquake, capable of Mw 9.0+. Produces the greatest vertical seafloor displacement and therefore the most destructive tsunamis. Cascadia, Nankai Trough, and Japan Trench are major megathrust sources relevant to insurance markets.
The maximum vertical elevation above sea level reached by tsunami water at the shoreline β the primary metric of tsunami coastal impact severity. Amplified by V-shaped bays and fjords (3β5Γ open-coast values), reduced on exposed headlands. The 2011 TΕhoku tsunami reached 40m run-up in Aneyoshi Bay, Iwate Prefecture.
The physical process by which a tsunami wave slows and increases in height as water depth decreases approaching the coast. Energy conserved from deep-ocean propagation is compressed into a shallower, slower wave with much greater height. The primary amplification mechanism transforming a sub-metre deep-ocean wave into a multi-metre coastal wave.
The horizontal distance inland reached by tsunami water from the shoreline. Governed by coastal topography, run-up height, and the gradient of the coastal plain. The 2011 TΕhoku tsunami penetrated 10km inland across the low-gradient Sendai plain. Steep coastal terrain limits inundation distance despite high run-up.
Probabilistic Tsunami Hazard Analysis β the tsunami equivalent of PSHA (Probabilistic Seismic Hazard Analysis) for earthquake. Integrates the full distribution of source scenarios, propagation models, and coastal amplification to produce probabilistic inundation maps at specified return periods. Most well-developed for Japan; used as the framework for PTVA (Probabilistic Tsunami Vulnerability Analysis) in commercial cat models.
Japanese Earthquake Reinsurance β the government-backed residential earthquake and tsunami reinsurance scheme in Japan. Private insurers provide the first layer; the government provides unlimited reinsurance capacity above specified thresholds. Triggered in 2011 TΕhoku, producing the largest JER payout in the scheme's history.
The geological record of past tsunami inundation preserved in coastal sediment sequences β sand sheets, marine diatoms, and organic material deposited inland by historic tsunami waves. Extends the tsunami record beyond the instrumental period (typically 150 years) by thousands of years. Essential for estimating return periods of large events on subduction zones with long recurrence intervals.
Erosion of soil from around building foundations by fast-moving tsunami inundation flow. Removes the bearing material supporting shallow foundations, causing settlement, tilting, and structural failure. A primary failure mechanism for buildings that might otherwise survive hydrodynamic loading β the foundation fails before the structure does. Pile foundations into bedrock or dense material are far more resistant to scour-induced failure.
Indian Ocean Tsunami Warning System β the warning and mitigation infrastructure established by UNESCO following the 2004 Indian Ocean tsunami. Provides seismic monitoring and tide gauge networks across the Indian Ocean basin to detect tsunamigenic earthquakes and issue warnings to coastal nations. Did not exist in 2004; its absence contributed significantly to the death toll.