RESEARCH DOSSIER
Digital Twin Infrastructure Indices for Parametric Insurance
Abstract. Digital twin infrastructure indices translate the recorded and reconstructed condition of physical networks into contractually specified contingent payments. This research investigates the scientific foundations of that translation across electric power, transportation, water, and coastal infrastructure. It distinguishes hazard intensity, physical asset damage, service degradation, economic loss, and financial liquidity need, each of which has a distinct measurement procedure and insurance significance. A digital twin can combine observations, physics-based simulation, network topology, exposure information, and probabilistic inference to estimate infrastructure states; the insurance value of its output depends on the reproducibility, timeliness, independence, and economic relevance of a specific index. The monograph examines the institutional origins of modeled-loss triggers, modern infrastructure modeling, observation and data standards, actuarial pricing, correlation, legal characterization, model governance, and validation. A transparent hypothetical electricity-network contract demonstrates how a service deficit index behaves under heterogeneous scenarios, alternative liquidity targets, and measurement failures. Comparative analysis of Caribbean parametric insurance, coastal flood modeling in Singapore, an illustrative Miami-Dade bridge network, tsunami-index design for Sendai Port, and Australian spatial twin initiatives delineates established applications from prospective insurance integrations. The central conclusion is that the incremental value of a twin must be established at the level of timely risk transfer under adverse conditions, with separate tests of physical fidelity, economic correspondence, settlement reproducibility, and contractual legitimacy. The study presents analytical synthesis and reproducible illustrative calculations without claiming a new named theory, empirical dataset, or demonstrated commercial adoption of a particular twin-based contract.
Keywords: Digital twin; infrastructure insurance; parametric insurance; modeled loss; index design; basis risk; service interruption; critical infrastructure; sensor fusion; contract verification; financial liquidity; actuarial modeling; network interdependence; catastrophe risk; model governance; uncertainty quantification.
Introduction
Critical infrastructure creates economic value through the persistent delivery of services. A bridge connects origins and destinations; a distribution grid makes energy available at particular locations; a water network supplies usable water at sufficient pressure and quality; a coastal barrier reduces the probability and consequences of inundation. The damage sustained by these systems is only one component of their financial and social consequences. Interruption, rerouting, emergency procurement, repair sequencing, lost tariff revenue, and public assistance can create different patterns of financial need. Insurance design must identify the obligation and beneficiary before treating any numerical representation of infrastructure as a measure of insured risk.
Parametric settlement promises a payment determined by the specified event and index. The NAIC's parametric insurance overview (2023) describes the use of predefined payment terms and designated verifiers. This institutional form can support rapid liquidity because the contracted calculation need not await full damage adjustment. Its speed depends on the availability of the agreed information, the legal conditions, certification, and the actual payment process. It does not follow from the presence of digital sensors alone.
The research object is an index derived at least partly from a continuously or periodically updated digital representation of an infrastructure system, combined with a contract specifying how a realized observation period determines a payment. This includes a modeled service deficit or modeled loss that is not identical to an independently assessed indemnifiable loss. A static catastrophe exposure model is a relevant antecedent; it qualifies as a digital twin for purposes of this study only where a specified relationship to changing physical asset or network states can be demonstrated. Definitions are functional and do not prescribe a vendor architecture.
The two governing scientific domains are Parametric and Index-Based Insurance and Insurance Technology. Their intersection concerns the conversion of measurement and computation into an enforceable settlement obligation. The present study concentrates on infrastructure-specific indices, independent evaluation, and the consequences of using dynamic asset representations in a financial contract. It does not reproduce the general science of index insurance, engineering inspection, or digital-twin risk prediction.
The investigation asks which infrastructure variables support reproducible, economically relevant settlement; how their errors propagate through payout rules; what empirical evidence distinguishes technical feasibility from financial performance; and how governance can preserve settlement integrity when the operational model changes. The method is a targeted critical review of institutional primary materials, official technical summaries, standards, and original research, followed by mathematical derivation and comparison of explicitly hypothetical scenarios. The literature search covers digital twins, bridge and flood networks, modeled-loss insurance, basis risk, data provenance, and actuarial modeling. It is not a systematic review with a registered screening protocol. No proprietary insurer policy wording, individual claims file, measured sensor series, or portfolio loss history is available; conclusions about implementation remain conditional on validation against such evidence. Source access is recorded at the level actually inspected.
Scientific Foundations and Historical Development
From asset representation to settlement index
Infrastructure modeling long predates the digital twin. Engineering drawings, geographic information systems, asset registers, finite-element models, hydraulic simulators, network flow calculations, reliability diagrams, and outage management systems each represent selected aspects of real systems. The digital-twin setting brings these representations into an update relationship with observations about identifiable assets. The scope can extend from one pump or bridge to a network of interacting systems.
The Australian spatial twin initiative described by ANZLIC (n.d.) illustrates an institutional conception of shared spatial representations for place-based planning and intervention testing. The initiative emphasizes cooperation and secure federation. The lesson for insurance development is that an index may depend on several organizations maintaining mutually compatible information about different parts of a system. Federation enlarges the set of possible measurements while distributing responsibility for data quality and continuity.
An information model records the identity, geometry, condition, and relations of assets. A physical process model predicts system response given inputs and boundary conditions. An observation model describes what instruments measure, including noise, bias, and coverage. A state estimator updates uncertain internal conditions from observations. A service model maps physical states to delivered output. A financial model maps consequences to expenditure, income loss, or liquidity. A contractual index selects and transforms one or more of these quantities. A sufficiently detailed geometry alone cannot establish the sequence; the connections between layers are the substantive scientific work.
For example, a bridge deck can remain undamaged while a flood closes approaches. Its load-bearing state, legal traffic availability, and contribution to regional travel may consequently diverge. A model tailored to material fatigue might estimate a useful latent engineering variable and still be unsuitable as an interruption settlement mechanism. A bridge operator facing emergency inspection costs, a logistics firm facing rerouting costs, and a municipality financing temporary access require different targets. The same physical event links these exposures but does not equalize them.
State, observation, and scope
Let X_t denote the physical and operational state of a defined network at time t. Let U_t contain exogenous inputs such as rainfall, wind, demand, and control actions. A generic state relation is:
X_(t+1) = F_theta(X_t, U_t) + W_t.
Observations Y_t satisfy:
Y_t = H_phi(X_t) + V_t.
Here F_theta is a transition representation, H_phi describes sensor response or other measurement processes, and W_t and V_t express process and observation discrepancies. These are conventional state-space expressions, not a proposed proprietary model. A twin can use deterministic or probabilistic estimation; the suitability of either depends on the physical system and available evidence.
The settlement index I_e for event e is a specified functional of observations, estimated states, asset identifiers, event windows, and admissible model versions. A forecast estimates possible future conditions. A contemporaneous estimate describes the event using information available by a specified cutoff. A retrospective reconstruction uses additional evidence. These can differ substantially. A contract must state which information set controls payment. Employing a later reconstruction to score historical forecasts may create an apparent accuracy that would not have been available in real settlement.
Four distinct objectives follow. Physical fidelity concerns agreement between modeled and independently observed states. Statistical calibration concerns the relationship of estimates or probabilistic predictions to outcomes. Economic relevance concerns correspondence to the policyholder's intended financial protection. Settlement reproducibility concerns the ability of authorized parties to obtain the same index and payment from the prescribed information. Success on any single objective leaves the others to be established.
Historical convergence with index insurance
Modeled-loss insurance already establishes a contractual relationship between prescribed models and financial payment. The CCRIF technical brief (CCRIF SPC, 2025) describes pre-agreed catastrophe models, hazard inputs, and attachment levels. Such mechanisms can be understood as predecessors to settlement based on asset-updated infrastructure models. They demonstrate the feasibility of specifying model-driven risk transfer; they do not establish that the facilities use continuous infrastructure twins in settlement.
The change introduced by a particularly dynamic digital representation concerns the refreshability of exposure and network information. A hazard-based index for a fixed location can remain stable throughout a policy term. A twin can incorporate a newly reinforced substation, changes in feeder connectivity, new bridge capacity, or changed flood defenses. This creates a tension between economic relevance and contractual certainty. More recent exposure information may improve prediction, while discretionary updates during coverage may alter the distribution of promised benefits. The technical ability to revise an index therefore requires contractual boundaries on the use of revised information.
Distinct infrastructure losses
Physical repair cost, replacement cost, revenue interruption, economic welfare loss, and liquidity requirement are different financial variables. A system operator might receive regulated revenue throughout an outage, incur significant emergency procurement expenditure, or face contractual penalties tied to service restoration. A business dependent on infrastructure can experience lost production without suffering any property damage. A government may incur response costs while the directly damaged asset belongs to a private operator.
The object of insurance must follow the identified financial interest. A service deficit can be a useful proxy for emergency expenditure with weak correspondence to the total social cost of a disaster. Conversely, a physical damage estimate can approximate repair expenditure while providing little information about interruption in a reroutable network. Any choice among these measures implies a distribution of payout across affected parties and levels of severity.
Primary Instruments and Digital Twin Measurement
Choosing a measurable service
Service indices can express the duration and intensity of lost supply. For an electricity network, aggregate unserved energy can be computed from the integral of expected baseline demand less delivered power, subject to a lower bound of zero. If power is measured in megawatts and time in hours, the index is megawatt-hours. A customer-duration index instead aggregates the number of customers interrupted across time, and division by a stated customer denominator gives an average interruption-duration measure. These variables can differ sharply in an industrial region with few large customers or a dense residential region with many small customers.
The EIA reliability table (U.S. EIA, n.d.) distinguishes reported distribution reliability measures by methodology, treatment of major event days, and loss-of-supply exclusions. Such distinctions matter directly for contracts. A policy intended to cover severe disruptions should not inadvertently adopt a reporting convention excluding the very events it seeks to finance. An event-specific insurance index also needs a declared time window and denominator rather than reliance on a yearly statistical aggregate.
For a road network, inaccessible route-hours, weighted trip delays, or lost origin-destination connectivity can describe deterioration in service. For water, the measure may be cubic meters of unmet demand, customer-hours below a stated pressure, or supply failing a defined quality criterion. For ports, loss of throughput relative to a baseline or berth closure-hours may better represent liquidity pressure than a regional hazard reading. For flood defenses, an exceedance of specified hydraulic capacity or modeled inundation depth at defined critical assets may be appropriate. Each requires a physical unit, spatial definition, temporal aggregation rule, severity threshold, and relationship to a financial objective.
A generalized service index for event e over [t0,t1] can be written:
I_e = integral from t0 to t1 of sum over i of w_i max(0, b_i(t) - q_i(t)) dt.
Here q_i is delivered service and b_i is the agreed baseline, in compatible units. Weights w_i can designate contracted exposure categories; if they are dimensionless they preserve the service unit, while monetary weights convert the index to a modeled monetary quantity. A contract must specify the meaning and normalization of the weights. If nodes represent the same service at different stages of production, summing deficits can double count loss. An electricity feeder and the substation that supplies it should not both be added as separate end-user deficits merely because the twin represents both.
Observations, metadata, and time
Insurance computation requires durable data lineage. An observation needs an identifiable sensor or source product, observation time, receipt time, spatial location, unit, calibration state, validity flag, transformation history, and the asset or geographic feature it represents. Station relocation, firmware replacement, sensor saturation, and changes in the sampling interval can produce discontinuities. An index based on threshold exceedance requires explicit treatment of readings at the boundary, rounding, and the duration represented by each sample.
The WMO observation guide index identifies technical chapters on measurement, observing systems and quality management, including testing and calibration (World Meteorological Organization, n.d.). This establishes the breadth of relevant metrological work, without establishing compliance of any sensor in an insurance system. Calibration documentation, traceability and quality flags must be examined for the actual instrument and measurement.
Interoperability can allow separate data providers to exchange observations and metadata. The SensorThings API overview (Open Geospatial Consortium, n.d.) describes Sensing and Tasking functions and support for heterogeneous observations. A common schema helps represent who or what reported a value, but data encoded in a valid schema can still be wrong. Semantic alignment additionally requires common interpretations of units, spatial reference systems, sampling and asset identity.
Time synchronization is particularly important during fast transients. Combining instantaneous power with a time-averaged demand baseline without alignment creates spurious deficits. Converting observations to a canonical time standard and preserving original timestamps enables reconciliation after daylight-saving changes, network delays or clock resets. A measurement arriving after the contractual cutoff must be governed by the late-data clause rather than accepted ad hoc.
Hazard fields and data fusion
Regional hazard products extend observation coverage beyond individual instruments. The ShakeMap product description (USGS, n.d.) establishes its use for near-real-time spatial estimates of earthquake shaking. Wald, Lin, and Quitoriano's ShakeMap uncertainty report (2008) explains the influence of observations and source geometry on spatial uncertainty and changes in confidence as information improves. A contract referring to a shaking map must identify product type, source authority, location interpolation, units and publication or revision cutoff. The first available map and a later revised map may produce different trigger values.
The Copernicus reanalysis description (Copernicus Climate Change Service, n.d.) describes gridded ERA5-family products, uncertainty information and preliminary availability. Such data can support historical calibration and fallback design where the coverage, scale and lag are appropriate. A reanalysis grid cell represents a spatially aggregated reconstruction; localized inundation near an embankment cannot be inferred reliably from its rainfall value alone. Any fallback needs validation against the primary index and an explicit transformation.
Combining observations and physics may reduce uncertainty about inaccessible assets. Nonetheless two apparent sources might share the same weather forecast, processing service or network connection. Treating dependent sources as independent can understate uncertainty. An index using a hydrological simulation, remotely sensed flood extent and local gauges should document whether the simulation assimilates the gauges or satellite product. Independent verification means independent evidence or independent control of the settlement process; requiring two copies of the same provider's estimate does not achieve either automatically.
A contractual data specification
An operational twin is ordinarily modified to improve asset management. The settlement environment should maintain a distinguishable, reproducible configuration. At inception, record the network boundary, asset identifiers, data provider obligations, model code and parameters, dependency versions, approved observation transforms, baseline rules, event definitions, rounding and limits. If input data are revised, the policy should specify which snapshot governs the calculation and whether a correction or appeal may alter it.
There are legitimate dynamic specifications. A contract may authorize an automated update of network topology at scheduled intervals using a signed register. The important feature is a deterministic admissibility rule known at inception, rather than an immutable representation of a changing network. An update can be restricted to future event windows, avoiding retroactive alteration of a settlement. An independent calculation agent should receive the same source snapshots and model version, subject to security restrictions.
The settlement archive should preserve the inputs, accepted and rejected observations, source quality flags, software identifiers, transformations, output, contract evaluation and certification. A cryptographic digest establishes whether stored bytes have changed. It does not establish that the original sensor reading was accurate or that a model was scientifically valid.
Measurement failures
Infrastructure failures can disrupt their own measurement systems. A destroyed meter, flooded communication room, fiber separation or electric-power failure may remove observations exactly when the insurance protection is most relevant. Missingness is consequently a possible consequence of the event. Replacing absent data with a normal reading without scientific justification turns severe adversity into apparent normality.
The IAIS index-insurance issues paper (2018, paras. 70–71) specifically recognizes credible calculation and contingency arrangements when extreme conditions damage measurement equipment or interrupt transmission. Fallback design should distinguish delayed transmission from destroyed observations; preserve a documented sequence of sources; prescribe deadlines, uncertainty handling, and dispute arrangements; and define what happens if all fallbacks fail. It should be tested under correlated observation loss rather than merely demonstrated under benign conditions.
Risk Transfer Mechanisms and Actuarial Construction
Payout functions and liquidity targets
Let L_e denote a realized eligible financial need, not necessarily the total physical or economic loss. Let T_e denote the target financing layer chosen at inception, for example a capped amount of emergency expenditure. Let I_e denote the certified index. A simple continuous payout schedule has attachment a, exhaustion b greater than a, and limit M:
P(I) = M max(0, min(1, (I-a)/(b-a))).
Its dimensions are coherent when I, a and b share units and M is expressed in money. The function is zero at and below a, linear between a and b, and equal to M at and above b. A stepped table, binary trigger, aggregate event limit or multiple-trigger condition can replace this schedule. CCRIF's modeled-loss attachment description (2025) supports the idea of predetermined attachment in an existing institutional program; the expression here is a generic illustration, not its actual policy formula.
A policyholder seeking immediate emergency funding may choose a target capped at available repair resources or administrative capacity rather than full damage. For example, an operator that needs $600,000 within five days to hire emergency crews may experience $4 million of final restoration cost. A $600,000 payout can satisfy its stated liquidity purpose while leaving large physical losses uninsured. Assessing the difference from total damage as model error would conflate contract structure and financing purpose.
Payment timing belongs to the objective. Let tau_e be settlement delay and d(tau) a factor that represents the economic availability of funds for the relevant response purpose. A simple effectiveness measure is d(tau)P, with d chosen transparently by the analyst. A payout that arrives after suppliers have withdrawn or emergency response has been delayed may provide lower useful protection despite matching the eventual monetary target. This dimension should be reported separately, since a discount factor cannot represent all nonlinear consequences of liquidity shortage.
Synthetic infrastructure example
Consider a hypothetical electricity distribution portfolio with an event window of ten hours. Define I as unserved energy over that window, based on a predetermined demand baseline after excluding planned interruptions. Define a = 200 MWh, b = 800 MWh and M = $600,000. The linear portion of the schedule pays $1,000 for each MWh above attachment. For clarity, all monetary values in the following table are in thousands of dollars; the observations, losses and probabilities are entirely synthetic.
| Scenario | Probability | Index (MWh) | Payout ($000) | Target liquidity ($000) | Shortfall ($000) | Excess ($000) |
|---|---|---|---|---|---|---|
| Mild disruption | 0.55 | 100 | 0 | 0 | 0 | 0 |
| Moderate interruption | 0.20 | 350 | 150 | 200 | 50 | 0 |
| Severe regional outage | 0.12 | 650 | 450 | 500 | 50 | 0 |
| Widespread outage | 0.08 | 900 | 600 | 600 | 0 | 0 |
| Costly local failure | 0.03 | 180 | 0 | 300 | 300 | 0 |
| High deficit, low expense | 0.02 | 500 | 300 | 100 | 0 | 200 |
Assume the six scenarios are mutually exclusive and exhaustive outcomes for a hypothetical single settlement opportunity. Probabilities sum to one. The expected payout is 0.20(150) + 0.12(450) + 0.08(600) + 0.02(300) = $138,000. Expected target liquidity is 0.20(200) + 0.12(500) + 0.08(600) + 0.03(300) + 0.02(100) = $159,000. Expected shortfall is $25,000 and expected excess is $4,000, so expected payout minus expected target equals negative $21,000, also excess minus shortfall. A payment occurs with probability 0.42; positive target occurs with probability 0.45. The probability of no payment given positive target equals 0.03/0.45, approximately 6.67 percent.
The examples identify two distinct problems. The costly local failure has expensive consequences despite a small aggregate service deficit, while a large service deficit can generate relatively low immediate expenditure under a different set of restoration conditions. A second metric for affected critical assets could help differentiate these cases, but adding that metric has consequences for measurement independence and price.
A sensitivity experiment lowers attachment to 100 MWh while leaving exhaustion and limit unchanged. The new payout for 350 MWh is approximately $214,286; for 650 MWh, $471,429; for 900 MWh, $600,000; for 180 MWh, $68,571; and for 500 MWh, $342,857. At exactly 100 MWh no payout occurs. The expected payout is approximately $156,343. Expected shortfall declines to approximately $10,371, while expected excess increases to approximately $7,714. The changes show why a lower attachment can reduce a particular measure of adverse basis risk while raising the cost of cover and expanding overpayment. Neither schedule is an empirically calibrated offer.
In the original schedule a severe event with true measured-equivalent index 650 MWh would pay $450,000. If communication failure causes the accepted index to be underestimated by 150 MWh, payment becomes $300,000. For a $500,000 target, the shortfall increases from $50,000 to $200,000. If instead a validated independent fallback yields 620 MWh, payment becomes $420,000. The comparison isolates the financial consequence of settlement observation error from the underlying economic mismatch. The numbers establish arithmetic possibilities, not estimates of actual fallback accuracy.
Asymmetric settlement objectives
A symmetric squared-error objective makes the conditional mean of a finite-variance financing target the optimal unconstrained point prediction under a given information set. If unmet need is more costly than an equal overpayment, the relevant optimization changes. Maier and Scherer's (2026) expectile analysis of basis risk derives conditional expectile solutions under asymmetric squared-error penalties. Their results belong to an explicit objective function and conditioning structure; they do not establish that one expectile level is appropriate for all policyholders.
For illustration, let t be a nonnegative random financing target, p a proposed payment, and alpha lie strictly between zero and one. Define the loss of the decision as alpha times the squared positive part of t minus p, plus one minus alpha times the squared positive part of p minus t. For an unconstrained optimum and finite second moment, differentiation gives alpha E[(t-p)+] = (1-alpha) E[(p-t)+], with both expectations conditional on the admissible information. At alpha = 0.5 the solution is the conditional mean. With larger alpha the objective gives greater emphasis to unmet need. Under a conditional two-state distribution with equally likely targets 0 and 100, the interior solution is p = 100 alpha: alpha = 0.8 gives 80 rather than 50. These arithmetic results are illustrative applications of existing theory.
The optimum of that decision loss need not satisfy affordability, regulatory, limit, monotonicity or premium constraints. A contract may set a limit of 60 in this example, forcing the optimum to 60. A fixed annual premium budget couples decisions across events because expected payments enter technical pricing. A cost comparison should test alternative weighting rules at the same budget and identify who bears any overpayment costs. For infrastructure, the choice of financing target is prior to optimization: a restoration cost, revenue deficit and emergency liquidity target generate different conditional distributions.
Steinmann et al. (2023) implement a parametric insurance design framework in CLIMADA and use stylized cyclone and winter-storm examples to quantify basis risk. An infrastructure-specific application could compare a hazard-only trigger and a twin-derived trigger on identical event sets, exposure inventories, target definitions and payout constraints. Their examples demonstrate a method of design and assessment; they do not establish the gain from live infrastructure telemetry.
Pricing and capital
For a specified policy period with N admissible events and event-level payout P_e, an aggregate limit A gives aggregate paid benefit min(A, sum of qualifying payouts), subject to event grouping and reinstatement terms. Expected aggregate payout is the actuarial starting point for a risk premium. Expenses, calculation-agent costs, data and model costs, taxes where applicable, capital cost, reinsurance terms, risk margin and profit may enter the charged premium. The mixture above is for one settlement opportunity; its $138,000 expected payout is not an annual premium unless the scenario probabilities describe the entire policy-year exposure under the stated contract.
The return distribution depends on joint hazard and exposure states. Modeling loss severity alone is insufficient where event frequency and persistence change the number of triggers. An event definition that partitions a three-day storm into three daily triggers can create dramatically different liabilities from a single occurrence definition. The model must match the contractual event clock and annual aggregate structure.
Infrastructure portfolios can exhibit common hazard, shared supply dependencies and shared technological dependence. Two insurers may believe they diversify across water and electricity when both use a common storm field, the same payment infrastructure or a common third-party settlement vendor. A reinsurance arrangement tied to a regional wind index may respond weakly to a cedent's service-outage cover if local flooding and supply failure dominate. The insurer-reinsurer mismatch is a distinct transfer-layer risk.
Simulations should combine plausible spatial hazard fields, the condition of assets and network connections, restoration resources, and failure of observation channels. Capital calculations require the distribution of aggregate payments and the tail of that distribution, including vendor failure scenarios. Economic pricing needs to recognize that data improvement can reduce some uncertainty while making the promise more sensitive to common model assumptions. Solvency conclusions require jurisdictionally applicable prudential methods and a portfolio; this hypothetical study supplies neither.
Multiple indices and incentives
A compound payout may use maximum, minimum, weighted sum, conditional or sequential functions of hazard and service indices. An AND condition can constrain payment to a covered external event plus a significant service interruption. It may also increase false negatives if one input fails. An OR condition can protect against several pathways of disruption while increasing payout frequency and potentially paying twice if separate limits are combined incorrectly. A weighted sum should combine commensurable normalized variables and should justify weights using validation and the client's financing objective.
The design also affects incentives. A purely external hazard index is comparatively insulated from the operator's restoration decisions. An observed interruption-duration index can be affected by how the operator allocates crews, reports service, or changes its demand estimate. A hazard-conditioned, independently measured service index can reduce some manipulation opportunities without eliminating them. Incentive assessment must consider the marginal payment relative to the cost and other consequences of delaying restoration.
Repair, maintenance, and adaptation can lower future exposure. Scheduled updates at renewal may support improved risk classification but require careful handling of affordability and incentives. Aligning contractual baselines with independently documented asset improvements can ensure that financial coverage evolves coherently with physical risk reduction. No claim is made that every infrastructure operator will respond in the same way.
Comparative Infrastructure Evidence
Caribbean modeled-loss insurance
CCRIF is the clearest institutional bridge between index insurance and complex catastrophe calculations in the selected evidence. Its April 2025 technical brief lists earthquake, tropical cyclone, excess rainfall and utility-related products and describes payments based on hazard information processed through catastrophe models. The report establishes an operational precedent for modeled-loss settlement and gives insurance-specific context for infrastructure exposures. It does not provide the complete signed utility policy wording, input governance rules, event-level data or a documented continuously synchronized asset twin. Those elements remain outside the examined material.
The relevant analytical lesson is that the contractual status of model output can be established independently of its exact technological origin. A twin-derived estimate would require the same prior agreement on how an index becomes payable, with added attention to dynamically changing exposure and network states. Revisions to a regional model and updates to an asset-specific twin need different governance intervals, because the latter may occur continuously.
Electricity and water utility products in operation
The May 2023 electric utilities product booklet explains how CCRIF's tropical cyclone calculation combines asset inventory, storm fields, engineering vulnerability and contractual parameters. The modeled exposure is overhead transmission and distribution infrastructure. Generation plants and substations are outside that model's exposure inventory, and the financial module does not model all economic activities dependent upon electricity. Attachment, exhaustion and the ceded share determine the coverage layer. Those boundaries matter: failure of a generation plant could produce a widespread deficit of electricity without generating equivalent modeled damage to covered lines. Conversely, damaged distribution assets could trigger a modeled-loss payment where network redundancy limits interruption. A service index and an asset damage index therefore address distinct economic interests even when attached to the same storm.
The operational record extends beyond the availability of a product. CCRIF's July 23, 2024 Beryl water utility payout announcement reports that Grenada's National Water and Sewerage Authority received US$2,201,833 following the hurricane of July 1. The insurer identifies the electric and water utility payments after Beryl as the first payments under those product lines. This is evidence of actual settlement rather than only a research proposal. It remains an insurer report: the examined record does not supply independent ground-up losses, signed wording, exact calculation inputs or a counterfactual restoration trajectory.
This evidence identifies an institutional starting point for incremental experimentation with twins. A utility could retain the hazard and physical-damage model for its repair-financing layer while separately testing a service-dependent measure for short-term continuity expenditure. The two measures would need separate eligible expenditure definitions to avoid confusing complementary financial protection with duplicate protection. A twin could inform both; a shared model would also create correlated model and data errors across the two layers.
Singapore and compound flood processes
In 2021 PUB announced the Coastal-Inland Flood Model to support joint simulation of coastal and inland flood risks and adaptation planning. Its later site-specific studies page describes using the model to locate vulnerable areas under rainfall and extreme sea-level conditions. This is documented primary evidence of combined hazard modeling for infrastructure planning in Singapore, not evidence of a particular insurance product.
Flooding illustrates why an index based on one meteorological variable may fail to capture the hazard experienced by an asset. The water level at a pumping station depends on local topography, drainage, river or coastal boundary conditions, antecedent water, and operational defenses. A coupled model could estimate inundation duration or service impairment at designated assets; such a quantity is an eligible research candidate for parametric coverage. Prospective calibration would need observed water levels, event reconstructions, outage or emergency expenditure records, and a defined set of admitted source versions.
Service interruption in compound flooding may occur at an apparently undamaged asset through access restrictions or electricity failure. A combined infrastructure index could account for those pathways if they belong to the covered peril and financial interest. Insurance use requires verifying the model under conditions that are rare even in long observational records, including simultaneous high rainfall and high sea levels.
Bridge-network digital twins
Geng and colleagues (2025) present a system digital twin framework for bridge-network life-cycle risk incorporating multiple data sources and Bayesian network dependencies. Their Miami-Dade illustrative example connects condition and traffic information with hydraulic observations. Their hypothetical network changes demonstrate an approach to updating a representation as its composition evolves.
The study provides a scientific rationale for expressing infrastructure risk at network scale. A bridge failure may be physically observable while its consequences depend on alternatives and shared river exposure. A route-specific interruption index could reflect users' access to alternative crossing points. A structural-condition index could instead address a specified emergency inspection or repair interest. These two indices may suggest quite different risk transfers.
Bayesian updating can combine evidence about uncertain component conditions and dependencies. A posterior failure probability is normally a risk estimate about possible outcomes rather than an observed failure. A contract paying when such a probability exceeds a threshold would be transferring a defined change in estimated risk, an unusual exposure requiring particular legal and economic justification. For immediate disaster liquidity, a model of realized network degradation or modeled event loss may be more interpretable. The examined research does not establish prospective payment accuracy, contract adoption, or insurer loss experience.
Sendai Port and tsunami indices
Miki and colleagues (2025) study tsunami parametric basis risk using probabilistic tsunami assessment and a Sendai Port application. The published abstract reports improved correspondence using inundation depth and multiple indices within its optimization exercise. The authors' funding acknowledgment identifies Swiss Re International SE, Japan Branch. The result is important evidence of research on local hazard indicators for infrastructure-related parametric design, but its performance remains specific to the study's scenario set, loss assumptions, payment structure and optimization procedure.
A relevant extension for a digital-twin setting would be to incorporate a maintained description of port assets, protection measures and service dependencies when estimating a contractually defined index. This is an analytical inference from the evidence, not a reported component of the cited study. Independent out-of-event and out-of-location evaluation would be necessary before generalizing any reduction in basis risk to other ports or hazards.
Spatial data governance in Australia
The ANZLIC spatial twin overview describes cooperation across institutions and securely shared representations. Insurers may have an interest in accessing consistent exposure descriptions for public and private assets, but availability of a spatial platform is not equivalent to permission to use its data for underwriting or settlement. Contractual access rights, security classification, commercial terms, source guarantees and revision histories determine applicability.
This distinction separates three forms of readiness. A city may possess an information-rich representation useful for planning; a model may be sufficiently validated for a particular engineering decision; an insurer may have a contract and independent computation capable of producing enforceable payments. The selected evidence supports the first two forms in particular contexts and supports modeled-loss insurance separately. It does not demonstrate an integrated, prospectively validated twin-trigger product across all of them.
Cross-case comparison
| Evidence | Documented object | Possible insurance relevance | Evidentiary boundary |
|---|---|---|---|
| CCRIF | Parametric modeled-loss settlement | Predetermined calculation and liquidity | Full utility wording and twin integration not examined |
| Singapore PUB | Coupled coastal-inland flood model | Asset-level hazard and service proxies | Planning and operational modeling, no observed insurance performance |
| Miami-Dade research | Illustrative bridge-network twin | Interdependency and network-level impact | Research demonstration; no policy outcomes |
| Sendai Port research | Optimized tsunami indices | Local hazard relevance and multimetrics | Scenario-specific research, no independent replication |
| ANZLIC | Spatial twin collaboration | Interoperable exposure information | No settlement or pricing record |
These cases establish complementary capabilities rather than one common deployment. Their synthesis provides a research pathway: use existing parametric settlement mechanics, infrastructure-specific hazard and state estimation, a service or financial objective, and institutionally credible observation arrangements. Demonstrating each capability in a different setting does not establish successful integration in a single contract.
Basis Risk, Error, and Alternative Explanations
Measuring adverse and favorable deviations
Define D = P(I) - T. The shortfall S = max(0, -D) is the unmet target; excess O = max(0, D) is the payment beyond that target. Expected shortfall E[S], expected excess E[O], conditional shortfall given a consequential event, and severe-loss conditional nonpayment probability each answer different questions. Expected D alone can obscure large adverse and favorable outcomes that cancel.
The IAIS discussion of basis risk (2018, Annex 1) distinguishes adverse outcomes with and without index response and emphasizes understanding the client's adverse event. The monograph adapts that distinction to measurable liquidity targets. A catastrophic loss above a knowingly agreed limit cannot be interpreted as measurement error merely because the amount beyond the limit remains unfunded. Both the contractual target and full ground-up loss should be reported where data allow, so layered risk transfer and unexpected mismatch can be distinguished.
Consider a portfolio in which a model predicts ordinary interruption accurately but misses rare multimodal failures. Mean squared error might favor that model over a simpler index because ordinary hours dominate the dataset. The insurer's payment schedule may operate almost entirely in the rare tail. A direct test of payout deviations, conditional on severity and timing, is a better measure of suitability for this application.
Decomposing index and target errors
Let Z denote a reference index derived from independently reconstructed service observations. Let I be the operational index available by the settlement cutoff. Define T as the intended financial target. Then
P(I) - T = [P(I) - P(Z)] + [P(Z) - T].
The first term measures the settlement consequence of the difference between operational and reference indices, including observation and implementation errors. The second term captures the mismatch remaining with the reference index, including payout schedule and economic correspondence. This is an exact algebraic identity. It does not require statistical independence. The reference index itself has uncertainty, so its reconstruction procedure and error interval should be stated.
For the continuous schedule in the synthetic example, the maximum slope is M/(b-a). Therefore, for two index values with identical admissibility conditions, the absolute payout difference is at most M/(b-a) times the absolute index difference. In the example that bound is $1,000 per MWh for the original schedule. A binary step has no analogous finite global slope bound: an arbitrarily small difference straddling its threshold can change payment by the entire step. Contract designers can use monotone continuous schedules to limit sensitivity to small measurement differences, while acknowledging that catastrophe losses and other eligibility clauses may still be discontinuous.
An observed adverse payout may arise because the event was excluded, because the index did not adequately proxy the target, because measurements were unavailable, because the model version was wrong, or because the contract was executed incorrectly. These mechanisms demand distinct remedial actions. Recalibrating a model cannot resolve a misunderstood exclusion; changing a payout schedule cannot fix a damaged timestamp service.
Credibility and validation
Shao, Hightower, and Schindel's digital twin credibility assessment (2023) identifies verification, validation and uncertainty quantification as important throughout the twin lifecycle. The source concerns manufacturing, and the present application transfers its methodological concepts rather than asserting that its industrial context establishes insurance validity. Verification asks whether the implementation conforms to the specified model and algorithm. Validation assesses adequacy for the intended purpose using independent observations. Uncertainty quantification characterizes limitations in inputs, assumptions and outputs.
A twin may be physically adequate for long-term maintenance yet inadequately validated for event-window payment. For a service index, validation should use independently assembled evidence of delivered service, not only the same telemetry that updated the estimator. For modeled financial need, validation also requires a comparable accounting definition and resolved treatment of deductibles, financing layers and timing.
ASB's ASOP No. 56 (2019, sections 1 and 3) provides a US actuarial professional context for understanding purpose, dependencies, assumptions, testing and governance. Its scope is materially relevant actuarial work under the applicable professional regime; it is not a universal legal requirement. The practical consequence for this subject is that an insurer evaluating outsourced infrastructure models must understand whether reliance is consistent with its intended use and quantify material model limitations.
A validation program should split data by whole events and network regions, rather than randomly splitting individual sensor observations that share strong temporal or spatial dependence. A replay evaluation must restrict each event to information actually available at the contractual cutoff. Results for ordinary and extreme events, different asset classes, urban and rural networks, communication failures and adaptation regimes should be reported separately. Sparse major-event records warrant explicit uncertainty and supplementary scenario stress tests, not claims of empirically established return periods.
The particular challenge of interconnected energy systems has been developed in Lonergan, Greco and Sansavini's (2023) energy infrastructure resilience perspective. They discuss the relationship of risk assessment, resilience and insurance design. Their argument supports the relevance of continuity metrics and multidisciplinary collaboration as research directions; it is not an empirical estimate of insurance-induced resilience.
A validation matrix should contain four independently interpretable cells: ordinary observed events, severe observed events, severe events with interrupted observations, and simulated events beyond the observed support. Each cell requires sample size, observation provenance, loss-definition consistency, reference-model uncertainty and agreement between reconstructed and contractually admissible inputs. Results from simulated events must remain identifiable as simulation. Tests should retain evidence both of false payouts and of missed adverse outcomes. One useful falsification exercise tests a baseline with a deliberately degraded subset of telemetry, then repeats the comparison when the lost nodes are selected because they were most heavily damaged. A random missingness assumption can otherwise make the simulation artificially benign.
Where a bounded service deficit is only partially recorded, an interval can expose the degree of unresolved uncertainty. Suppose a 100 MW contracted load is evaluated over four hours; measured unserved energy totals 60 MWh over three valid hours, while the remaining hour has no observation. Under the declared physical bound of zero to 100 MWh for that hour, total deficit lies within 60 to 160 MWh. For an illustrative linear payout from zero at 80 MWh to US$400,000 at 160 MWh, the possible payout range is zero to US$400,000. No point estimate is justified by these bounds alone. A contractual fallback could use independent service logs or a pre-agreed conservative value; evaluating that fallback requires the distribution of missingness during adverse events. The interval itself is a diagnostic, not a proposed settlement rule.
Uncertainty, measurement selection, and nonstationarity
The relevant uncertainty includes aleatory variability in hazard, asset response and restoration; epistemic uncertainty about the model and parameters; measurement uncertainty; and contractual implementation uncertainty. Their boundaries vary with the estimation method. A simulator's output distribution can understate economic uncertainty when recovery cost variation or upstream supply dependencies are omitted. More detailed component geometry may reduce one error source while exposing sensitivity to poorly characterized parameters.
Nonstationarity creates further difficulty. Climate, exposure, maintenance, network configuration, land use and technology can change through time. A historical loss series from an older network is not automatically representative of the modern insured system. Historical reconstruction should state whether a historical hazard is applied to the historical or current asset representation and for what purpose. The former supports understanding historical loss correspondence; the latter supports a counterfactual stress or pricing scenario. Conflating them can bias calibration.
A counterfactual baseline also has uncertainty. Real demand during an outage is unobservable at affected endpoints. Weather-adjusted demand forecasts, recent unaffected reference days and comparable feeders can generate alternative estimates. Their performance in normal periods is testable, but performance during extreme events may be affected by evacuation, business closure, price changes and deliberate curtailment. The contract should predefine the baseline and avoid choosing the most favorable baseline after the event.
Security and independence
Voas and colleagues (2025) identify cybersecurity and trust issues in their NIST digital twin report. The insurance implication is that manipulation of observations or control-state information may change payment. Exposure includes replayed sensor messages, forged timestamps, unauthorized changes to topology and parameters, compromise of the calculation agent, and denial of availability. Confidential infrastructure information may also be disclosed through overly detailed reports.
Controls should separate data capture, model administration, payout certification and financial authorization. Access should be limited to the minimum needed, recorded in independent logs, and governed by change control. Historical data and model artifacts should be retained in a manner compatible with lawful access, infrastructure security and contractual dispute periods. A signed observation establishes its provenance relative to a signing key, not the physical accuracy of the sensor.
Independent evaluation has an economic dimension. The insured operator may control important service data, the insurer may control the payout model, and a vendor may be paid by one of the parties. Contractual designation of an independent expert, reproducible calculation, audit access and a correction process can address these conflicts. A data provider's technical expertise alone does not settle the question of financial independence.
Differential protection
Service metrics distribute effective coverage through their weighting rules. An unweighted customer-duration index treats customers equally in the calculation but may obscure differing medical, industrial or municipal consequences. An energy-weighted index gives large loads greater influence. A critical-facility weighting scheme directs more protection toward particular public-interest functions and requires transparent justification. These are choices about the insured objective, not neutral properties of telemetry.
Coverage validation should identify regional differences in data quality and distributional effects. A rural area with weak communications may have greater service risk but lower measurement availability. If an index simply excludes low-confidence areas, it can concentrate protection in already better-instrumented communities. A more expensive redundant observation system or carefully validated regional fallback may improve availability; its cost must be explicitly included in the assessment of insurance value.
Integrated Scientific Synthesis and Research Agenda
Contractual and legal architecture
The IAIS 2018 issues paper identifies legal certainty, insurable interest, credible indices and consumer protection as important matters for indexed insurance. Applicable insurance law differs by jurisdiction and line of business. Whether a modeled service-loss contract is classifiable as insurance, a financial derivative, or another arrangement depends on applicable law, covered interest, event, payment conditions and regulatory characterization. The paper does not resolve the legal status of every infrastructure product.
Economic design must be linked to the entity that experiences the relevant adversity, such as the asset owner, service operator, concessionaire, dependent business or public authority. Where multiple parties bear overlapping burdens, contractual allocation and coordination with other covers are needed. A payment calculated through a service index may coexist with property indemnity, contingent business interruption insurance or disaster reserves. Conditions for offsets, duplication and use of proceeds depend on actual contract wording and legal rules; they cannot be inferred solely from the index.
A complete specification should state covered peril and interest, event occurrence and grouping, geographic and asset scope, period, observations and their data rights, certified index formula and units, permitted model and update procedure, baseline, attachment, exhaustion, limits, verification and fallback, payment deadline, treatment of correction, expert determination and dispute resolution. It should clarify whether payment requires evidence of an eligible loss or other conditions beyond index response. The appropriate drafting approach varies with jurisdiction and the nature of the insured institution; this study supplies scientific design considerations and no legal opinion.
Operational settlement
A reproducible event process begins with registering an event candidate and locking its information boundaries. Source observations are ingested and validated against the agreed data protocol. The permitted model state and transformations are selected. The calculation agent produces a certified index and a concise report of the calculation. The insurer verifies contractual conditions, applies the payout schedule and authorizes payment. Any correction or challenge follows the prescribed procedure. A parallel record of actual financial outcome supports future evaluation but need not control payment when the applicable coverage does not require it.
The NAIC verifier discussion (2023) supports the relevance of designated measurement authorities and contingency arrangements. An insured receiving a zero determination should be able to distinguish a genuine below-threshold index from unavailable observations and computational failure. A claim notice, certified index result, payout explanation and path to correction should be traceable to an event identifier.
Settlement speed is a measurable outcome. Track delay from event end or qualifying trigger to data availability, certification, authorization and cleared transfer separately. An average across all events can conceal unacceptable latency for the largest events. Report the upper quantiles of delay, the frequency of fallback invocation, unresolved data gaps and frequency of disputes. Retain anonymized reports where sharing is appropriate.
Decision to adopt a twin-derived index
The scientific evaluation should compare against simpler feasible alternatives: a regional hazard trigger, a local observation trigger, a static modeled-loss estimate, and an infrastructure-state or service index. Comparisons must hold the target, covered events, limits and financial budget sufficiently constant to distinguish the effect of the index from the effect of buying a greater amount of coverage.
A useful twin adds incremental information about the intended financing loss or materially improves availability, timeliness, or contestability. The relevant gains should be weighed against additional sensor operation, model upkeep, independent verification, cyber defenses, licensing, redundancy, and actuarial uncertainty. If a simpler trigger achieves comparable protection at lower cost and with greater robustness, twin integration has little demonstrated incremental economic value for that specific product. This is a product-specific analytical conclusion rather than a universal rejection of complex indices.
The most promising candidates under this analysis are systems where network dependencies or localized physical states materially shape liquidity need and where sufficiently reliable observations exist. A local port or multi-feeder utility may satisfy these criteria; each requires empirical testing. A diffuse public-health or economic impact target may depend on institutions and recovery decisions outside the modeled asset network, creating large residual basis risk even when the twin accurately represents physical stress.
Research program and reproducible assessment
A prospective evaluation should preregister the definition of adverse outcome, the event window, severity strata, threshold family, baseline procedure, data freeze, and acceptance metrics. Collect hazard inputs, independent asset states, service data, restoration expenditure, insurance payments and settlement timestamps. Define an equilibrium between the historical reconstruction horizon and changes in exposure. Distinguish data missing because of routine maintenance from loss correlated with the covered peril.
Develop a baseline trigger and candidate twin-derived trigger using training events. Preserve held-out events and regional systems for external validation; avoid selecting weights using outcomes reserved for evaluation. For rare catastrophes combine transparent stress scenarios with empirical information and state where scenario probabilities come from. Report uncertainty around false-negative rates, shortfall and premium estimates; small samples do not support precise claims about rare-tail probabilities.
Compare the candidates at matched expected payout or matched premium, with an explicit account of operating expenses and risk margin. Test sensitivity to spatial resolution, data loss, event boundaries, alternative baselines, common-service failure, policy limit, exposure updates, and dependence structure. Include tests that independently reproduce payment from archived inputs. Institute an approved change process and track whether model revisions improve physical fit while altering payout frequency or distribution.
Field studies should distinguish technical accuracy from financial welfare. Useful outcomes include restoration procurement times, emergency borrowing, interruption duration, distribution of benefits, renewal and disputes. Observational differences between insured and uninsured operators may reflect size, credit quality, governance, risk management, or geographic selection. Causal evaluation would need a credible identification design such as randomized phases where feasible or carefully justified quasi-experimental comparisons. No causal effect on resilience is established by the cases reviewed here.
Original analytical contribution and boundaries
The original synthesis of this monograph is the integration of four evaluation domains: the physical state of infrastructure, the statistical behavior of an index, correspondence to an expressly defined financing target, and the reproducibility of the contractual calculation. These dimensions are familiar in their respective sciences; their combination clarifies why success of an engineering twin cannot by itself establish insurance suitability.
The decomposition of payout error shows precisely how measurement and model discrepancies enter payment separately from the financial mismatch remaining under a reference index. The continuous schedule's slope bound relates engineering accuracy requirements to monetary sensitivity. The hypothetical table shows the effect of alternative thresholds and observation failure on a liquidity target. These are applications of elementary mathematical and actuarial methods, not newly claimed theories.
The comparative study finds documentary support for a sequence of capabilities: catastrophe models can determine contractual payouts; infrastructure models can incorporate relevant dependencies and compound hazards; localized modeled indices can be optimized against simulated loss outcomes; secure data and governance arrangements are necessary for credible computation. Evidence for a mature, generalizable digital-twin infrastructure insurance market would require primary contracts, settlement records, and independent prospective validation. The work does not infer that evidence from technological analogy.
Conclusion
Digital twin infrastructure indices are a technically plausible instrument for structuring parametric risk transfer when they describe an infrastructure state or service that corresponds to an established financial interest. They can incorporate information about network connectivity, changing exposure, compound hazard and operational performance that a simple regional hazard index may omit. Their economic usefulness is conditional on the ability of a specific contract to obtain reliable observations, calculate a stable index and deliver resources in the relevant adverse states.
The key scientific distinction is the separation of actual harm, estimated infrastructure state, index value and contractual payment. Each involves a different evidentiary and analytical task. Direct integration of observations improves some forms of physical inference while introducing new dependencies. A more detailed model may reduce basis risk for a precisely identified exposure and increase fragility if the required data fail during a catastrophe.
The study's synthetic analysis illustrates why both shortfall and excess must be measured alongside expected payout, why changing attachment affects the cost-protection trade-off, and why observation error close to a settlement boundary can materially alter financial outcomes. Comparative evidence supports the constituent methods and modeled-loss precedent, but does not establish the insurance performance of any specific integrated digital twin. That performance remains an empirical question for the intended network, contract, jurisdiction and population.
The sound direction for research and product design is disciplined comparison with simpler triggers, independently validated event data, explicit settlement cutoffs, authorized version control, transparent service and financial targets, robust fallback calculation, and portfolio-level accumulation analysis. The value of digital infrastructure modeling in insurance is realized through its contribution to credible, economically relevant and timely protection. Scientific validation of that contribution is the necessary foundation of any decision to institutionalize it.
References
Actuarial Standards Board. (2019). Modeling (Actuarial Standard of Practice No. 56; effective October 1, 2020). Source.
ANZLIC. (n.d.). Principles for spatially enabled digital twins of the built and natural environment in Australia. Retrieved November 18, 2026. Source.
CCRIF SPC. (2023, May). CCRIF SPC's electric utilities product: Providing financial protection for Caribbean electric utility companies. Source.
CCRIF SPC. (2024, July 23). CCRIF's newest parametric insurance product – for the water utility sector – tested by Hurricane Beryl: Grenada's water utility receives US$2.2 million (EC$5.9 million). Source.
CCRIF SPC. (2025). Brief on CCRIF SPC (July 2024, revised April 2025). Source.
Copernicus Climate Change Service. (n.d.). Climate reanalysis. Retrieved November 18, 2026. Source.
Geng, Z., Zhang, C., Jiang, Y., Pugliese, D., & Cheng, M. (2025). Integrating multi-source data for life-cycle risk assessment of bridge networks: A system digital twin framework. Journal of Infrastructure Preservation and Resilience, 6, Article 9. doi:10.1186/s43065-025-00121-7. Source.
International Association of Insurance Supervisors. (2018). Issues paper on index-based insurances, particularly in inclusive insurance markets (approved May 25, 2018). Source.
Lonergan, K. E., Greco, S. F., & Sansavini, G. (2023). Ensuring/insuring resilient energy system infrastructure. Environment Systems and Decisions, 43, 625–638. doi:10.1007/s10669-023-09928-9. Source.
Maier, M. J., & Scherer, M. (2026). Expectiles as basis risk-optimal payment schemes in parametric insurance. European Actuarial Journal, 16, 3–36. doi:10.1007/s13385-026-00447-w. Source.
Miki, Y., Chua, C. T., Suppasri, A., Cheng, A.-C., Iwasaki, T., Shinozuka, Y., Ogawa, T., & Imamura, F. (2025). A proposed approach towards minimizing basis risk in tsunami parametric insurance scheme. npj Natural Hazards, 2, Article 67. doi:10.1038/s44304-025-00127-x. Source.
National Association of Insurance Commissioners. (2023, December 21). Parametric disaster insurance. Source.
Open Geospatial Consortium. (n.d.). OGC SensorThings API standard. Retrieved November 18, 2026. Source.
PUB, Singapore's National Water Agency. (2021, April 9). PUB commences development of Coastal-Inland Flood Model. Source.
PUB, Singapore's National Water Agency. (2026, March 3). Site-specific studies. Source.
Shao, G., Hightower, J., & Schindel, W. (2023). Credibility consideration for digital twins in manufacturing. Manufacturing Letters, 35, 24–28. doi:10.1016/j.mfglet.2022.11.009. Source.
Steinmann, C. B., Guillod, B. P., Fairless, C., & Bresch, D. N. (2023). A generalized framework for designing open-source natural hazard parametric insurance. Environment Systems and Decisions, 43(4), 555–568. doi:10.1007/s10669-023-09934-x. Source.
U.S. Energy Information Administration. (n.d.). Electric Power Annual: Table 11.1. Reliability metrics of U.S. distribution system. Retrieved November 18, 2026. Source.
U.S. Geological Survey. (n.d.). ShakeMap. Retrieved November 18, 2026. Source.
Voas, J., Mell, P., Laplante, P., & Piroumian, V. (2025). Security and trust considerations for digital twin technology (NIST IR 8356). National Institute of Standards and Technology. doi:10.6028/NIST.IR.8356. Source.
Wald, D. J., Lin, K.-W., & Quitoriano, V. (2008). Quantifying and qualifying USGS ShakeMap uncertainty (Open-File Report 2008-1238). U.S. Geological Survey. Source.
World Meteorological Organization. (n.d.). Guide to instruments and methods of observation (WMO-No. 8): Knowledge Hub. Retrieved November 18, 2026. Source.