RESEARCH DOSSIER
Unmanned Aircraft Insurance and Third Party Risk
Abstract. Unmanned aircraft insurance is the contractual and institutional financing of physical loss and legal liabilities arising from remotely piloted and increasingly automated aircraft operations. The principal scientific difficulty is that operational safety, responsibility for harm, and insurance coverage are determined through different evidence and decision processes. A drone may complete a lawful flight and still injure an uninvolved person; a breach of operating rules may establish important facts without, on its own, resolving civil liability or the effect of a policy exclusion; and an insured aircraft loss may arise from the same occurrence as a substantial uninsured third-party exposure. This dossier examines the development of drone insurance, operator eligibility, ground and air risk, policy terms, claims causation, autonomy, underwriting, actuarial quantification, and international legal variation. Primary evidence comprises European Union aviation and insurance instruments, 2026 European Aviation Safety Agency guidance, the United Kingdom Civil Aviation Authority's insurance rules, United States Federal Aviation Administration operational and reporting material, Canadian and Australian aviation guidance, and an Australian Transport Safety Bureau collision investigation. The contractual analysis examines the International Underwriting Association's AVN146 and AVN157 model-policy publications and their specialized endorsements. Peer-reviewed quantitative studies are assessed according to their units, assumptions, uncertainty, and ability to support insurance pricing rather than safety authorization alone. An illustrative frequency–severity calculation demonstrates why low occurrence probabilities cannot be interpreted without geographic concentration, conditional injury and damage, policy limits, defense expenses, and correlated causes. The comparative findings establish that insurable exposure is produced jointly by mission design, aircraft and control-system reliability, persons and property at risk, civil-law allocation, policy structure, and recoverability. The work provides an evidence-grounded assessment of the scope and limits of currently observable third-party drone risk, with an agenda for exposure-based loss data and transparent policy interpretation. It advances analytical conclusions without proposing a new theory, doctrine, or proprietary actuarial model.
Keywords: Unmanned aircraft; unmanned aircraft systems; UAS; drones; remotely piloted aircraft; operator liability; third-party injury; property damage; aviation insurance; aircraft hull insurance; civil liability; SORA 2.5; autonomous flight; beyond visual line of sight; risk assessment; actuarial frequency and severity; insurance exclusions; occurrence evidence; product liability; command-and-control link; aggregate exposure; insurance regulation; comparative aviation law.
Introduction: Subject, Method, and Research Questions
The scientific object of unmanned aircraft insurance is the conversion of aircraft-related uncertainty into defined contractual obligations. The insured exposure may concern a drone's physical structure, its payload, the operator's legal liability to injured people, damage to another person's property, or claims arising from collision with another aircraft. A single occurrence can activate different risk relationships: the aircraft owner loses an asset, the operator may owe compensation, a manufacturer may face a product-defect allegation, and a landowner or client may assert contractual rights. These interests require separate identification before a loss can be classified, estimated, or insured.
This dossier uses unmanned aircraft for the airborne vehicle and unmanned aircraft system (UAS) for the vehicle and associated elements such as the control station and command-and-control link. An operator is the person or legal entity organizing or conducting the operation under the applicable rule; a remote pilot is the person carrying out the defined piloting functions, where such a role exists. The policyholder, named insured, aircraft owner, contractor, remote pilot, and operating entity may be distinct. Third-party risk concerns adverse consequences imposed on persons or property outside the relevant insured interest, subject to the definitions of the policy and governing law. Injury, damage, a tort claim, an insured claim, an expense payment, and a regulatory occurrence report each form a distinct observation.
The research method combines primary-instrument reading, jurisdictional comparison, published model policy analysis, an event reconstruction using a public investigation, and actuarial analysis of exposure and loss. Its empirical claims are limited to the content of the identified sources. Numerical scenarios below are transparent illustrations, without proprietary claim observations or an estimated market loss ratio. The effective-state comparison is taken as of November 17, 2026. Aviation requirements and insurance duties may change; the actual date, category, aircraft, mission, and jurisdiction govern any real assessment.
Three central questions organize the work. First, how do law and operational records establish eligibility to fly, and how is that eligibility evidenced to an insurer? Second, which mechanisms translate loss of control, collision, human injury, or property damage into legal liability? Third, which events trigger a drone insurance agreement, and how can analysts evaluate the frequency, severity, dependence, and evidentiary uncertainty of the resulting claims? The distinction between these questions is reflected in the European regulatory emphasis on operational categories and risk mitigation in Regulation (EU) 2019/947 and in the independent treatment of aircraft liability insurance under Regulation (EC) 785/2004.
Conceptual and Historical Foundations
Unmanned flight as an insurance subject
Insurance institutions originally developed aviation liability around a person-controlled aircraft, a relatively visible pilot hierarchy, conventional maintenance responsibilities, and recognizable air-carrier operations. Civil drones broaden the operational range from recreation to photography, infrastructure inspection, surveying, agriculture, emergency support, logistics, and increasingly autonomous services. The material insurance issues evolve with the mission. A small camera drone flying in a controlled rural worksite and a delivery aircraft following scheduled urban routes may have similar equipment values yet markedly different third-party loss distributions. The latter presents higher populations exposed per operating hour, different collision and descent pathways, and a larger potential for correlated incidents if many units share software, power systems, fleet-management platforms, or airspace corridors.
The institutional shift is visible in the Aviation Insurance Clauses Group's publication of AVN146 for commercial UAS in 2024 and the model AVN157 unmanned-aircraft policy in 2025. The 2025 announcement reports a product combining physical damage and third-party legal liability, with optional endorsements addressing unauthorized use, breach of aviation regulations, supplementary payments, medical expenses, trespassers' costs, expanded aviation liability, extended hull risks, and cargo liability. These documents establish the availability of model wording and modular coverage design; they do not establish that every insurer automatically incorporates every endorsement or that an individual policy has a particular coverage outcome.
Layers of risk and the unit of analysis
An operating mission generates at least five analytically distinct layers. Technical hazard concerns malfunction, command-link degradation, navigation error, adverse weather, battery or propulsion failure, collision with an obstacle, or collision with aircraft. Physical consequence concerns where a failed vehicle comes to rest and the energy transferred to a person or object. Legal consequence concerns duties, causation, available defenses, and responsibility shared among persons or companies. Contractual consequence concerns the insuring agreement, definitions, conditions, exclusions, limits, deductibles, and endorsements. Financial consequence concerns claimant recovery, defense expense, insurer payment, reinsurance, and insolvency risk. Every transition requires an evidentiary step.
The elementary exposure unit is often one flight hour, but a flight hour alone omits the number and vulnerability of people beneath the route. Other potentially informative units are missions, kilometers flown, aircraft-days, aircraft-hours above populated cells, takeoffs and landings, geofenced operating-hours, or deliveries. A modeled risk per flight hour derived from a low-density corridor cannot be transferred unchanged to a crowded street. Historical flight counts are also vulnerable to selection: operators who report detailed telematics may be safer or more professional than operators who do not. The research therefore distinguishes hazard frequency from insured claim frequency and requires an explicit exposure denominator.
A major technical source, la Cour-Harbo's ground-impact fatality assessment, models uncontrolled descent across a route using probabilistic impact location and operational conditions. The paper demonstrates how such risk depends on the mission path and distributional assumptions. Its method is relevant to technical exposure estimation, while a complete insurance expected-loss estimate additionally requires the distribution of bodily injury, property damage, legal liability, defense expense, and coverage.
International institutions and limits of equivalence
The International Civil Aviation Organization's RPAS Manual, Doc 10019 addresses integration of remotely piloted systems into airspace, including command-and-control links, detect-and-avoid capabilities, operations, and legal and security issues. It is a technical and institutional reference rather than a uniform global drone insurance contract. The ICAO Annex 13 investigation framework has an accident-prevention objective, and the use of an investigator's causal finding in insurance litigation requires separate legal analysis. Aviation safety investigations and contractual claim adjustment may depend on the same physical evidence yet apply different tests.
Cross-border comparison must retain the legal category of the mission. European rules distinguish open, specific, and certified operations; United States small-UAS rules operate through Part 107 and associated conditions; the United Kingdom has its own insurance requirements; Canada's aviation authority describes public liability insurance as recommended rather than generally required for its drone rules; and Australia provides operator licensing, registration, approval, and insurance-check guidance. These are jurisdiction-specific observations. A single international statement that all drones must carry the same insurance would erase material differences in scope and interpretation.
Operator Eligibility and Regulatory Evidence
The European Union: operational authorization and compulsory cover
Under Commission Implementing Regulation (EU) 2019/947, operational obligations depend on the category and circumstances of the flight. Open-category operations are subject to mass, distance, visual-line-of-sight, airspace, and operational limitations. Flights outside the open-category conditions generally enter the specific-category authorization or declaration framework, as applicable. Risk assessment and mitigation are central to specific-category authorization. Insurance evidence may be required in an authorization application when Union or national law demands cover. The legal condition for safe operation remains analytically separate from the terms of insurance bought by the operator.
The insurance baseline is established in Regulation (EC) 785/2004, which addresses minimum insurance for aircraft operators and contains scope exceptions, including model aircraft with a maximum takeoff mass below 20 kg. For applicable aircraft in the lowest mass class under Article 7, the specified minimum for third-party liability is 0.75 million special drawing rights (SDR) per accident; the applicability of the regulation to a particular drone must be established before quoting that minimum as a legal requirement. The EASA insurance guidance states that operators of drones above 20 kg are required to have insurance and that most EASA Member States also require third-party liability cover for lighter drones. National requirements, the recreational/commercial distinction where relevant, and the specific scope of insurance obligations must be checked separately. There is no universal inference that every sub-20-kg flight in all European states is either compulsory or exempt.
The 2026 European implementation context incorporates SORA 2.5 guidance through EASA's September 2025 amendment, documented in ED Decision 2025/018/R. SORA evaluates third-party ground and air risk and structures assurance required for authorization of operations in the specific category. Its purpose is operational safety assessment. A SORA output may inform underwriting, but neither an authorization nor a quantified safety target is a promise that a civil claim will fall within a policy.
United Kingdom: the purpose of use matters
The UK Civil Aviation Authority's current insurance requirements for drones provide that drone use for a commercial purpose requires third-party insurance at a minimum, while recreational use below the relevant weight threshold may be optional. The CAA directs operators to cover compliant with the UK's assimilated insurance regulations. Its guidance for working with drones also distinguishes registration, remote-pilot competence, applicable operational authorization, and insurance. An organization must reconcile its registered operator identity with the insured entity: a certificate issued to one corporate company cannot be assumed to cover a different contractor simply because both participate in the same job.
Insurance verification in a UK commercial operation should therefore identify the operator ID, the actual operator's legal name, the work performed, the flight area, the insured aircraft, permitted pilots, the insurance period, and the policy's territorial and activity conditions. The CAA specifically advises that the insurance policyholder name match the name shown on the operator ID. This documentary discipline prevents an appearance of compliance created by a certificate whose named insured or coverage class is unrelated to the operation.
United States: operational regulation and insurance procurement
The Federal Aviation Administration's small-UAS Part 107 overview describes the federal operational regime for qualifying civil drones and includes pilot qualification, aircraft registration, permitted flight conditions, and airspace authorization. The cited overview does not establish a general federal third-party insurance obligation for all Part 107 operators. Contractual requirements imposed by clients, venues, or counterparties, and any independently applicable state or local duties, require separate identification. The absence of an insurance command in a general federal flight rule does not preclude operator liability for negligent or otherwise actionable conduct.
For evidentiary purposes, the FAA accident-reporting guidance requires a report within ten days when the specified serious-injury or loss-of-consciousness threshold is reached or qualifying property damage exceeds $500, with the valuation rule stated by the FAA. A report threshold is not an insurance deductible, an injury definition for all civil actions, or a complete database of insurable incidents. Remote identification can assist event reconstruction: the FAA's DiSCVR explanation describes how identification data can help exclude or identify candidates during an investigation, while emphasizing that an absence of data is not proof of unlawful flying. Identification technology supports provenance, not automatic assignment of fault.
Canada and Australia: contrasting institutional emphasis
Transport Canada's drone-pilot best practices expressly recommend public liability insurance and say that it is not required under the general Canadian drone framework described in the guidance. The authority also notes that standard household insurance may not respond to drone use. This confirms that recommendation, legal compulsion, and actual insurance availability are separate propositions. Insurance may still be required by a client, a property owner, a particular contract, or another applicable rule.
Australia's Civil Aviation Safety Authority advises those hiring a drone operator to verify registration, licensing or accreditation, required approvals, and the scope of the operator's insurance. CASA's advice that illegal operation could affect insurance response is a reason to inspect the contract, rather than assume every breach causes automatic avoidance. The same authority's separation of flying credentials and insurance evidence is especially important for outsourced survey, film, agricultural, utility, and infrastructure missions.
An evidentiary chain for eligibility
Operational eligibility is a time-specific factual claim. A defensible underwriting record should pair the operator's identity and authority with proof of aircraft registration where applicable, pilot competence or supervision, mission purpose, approved geographical and altitude limits, equipment designation, training, maintenance and preflight records, and effective insurance documentation. Authorization instruments are meaningful only within their dates, conditions, and jurisdiction. A certificate of insurance confirms what its issuing document actually states; it is normally a summary rather than a substitute for the signed insuring agreement and endorsements. When a mission changes from rural visual-line-of-sight inspection to urban beyond-visual-line-of-sight delivery, the record must be reconsidered rather than automatically carried over.
Third-Party Harm and Allocation of Responsibility
The physical geography of loss
Ground impact is the most visible potential third-party event, but the insured risk is wider. A collision with an occupied building may fracture glass, injure a person inside, damage expensive equipment, and interrupt a business. A collision with a power distribution component may damage property and produce consequential claims. An aircraft-to-aircraft collision can generate aviation liability of a far larger order than the insured drone's hull value. A delivered payload may be damaged or cause harm after release, raising questions about cargo, products, and contractual liability. The mechanism of injury must be traced before these outcomes are aggregated in a claim.
For a given route, collision consequences depend on aircraft mass, speed, geometry, propeller and battery hazards, impact material, the probability of a failure mode, the distribution of descent points, and the number and protection of persons or assets beneath those points. The article Quantifying Risk of Ground Impact Fatalities by la Cour-Harbo (2019) demonstrates a spatial stochastic approach rather than treating all flight hours as physically interchangeable. Jiang, Blom, and Rattanagraikanakorn (2025) investigate the relationship between an impact area, fatality consequences, and UAS design in their joint impact-and-fatality study. These analyses identify important engineering mechanisms; their results depend on modeled exposure, susceptibility, and operational assumptions. A probability of fatality is not a general prediction of the monetary value of all third-party claims.
A further ground-impact uncertainty study (2021) emphasizes that uncertain descent dynamics affect the distribution of ground impact locations, with horizontal drift and meteorological variation influencing where harm occurs. A national-airspace safety assessment (2018) examines both ground impacts and midair collisions, recognizing that population, shelter, aircraft density, and system reliability contribute to overall risk. Insurance analysis must convert these mechanisms into expected monetary liability and its tail, a step that may require data unavailable from aviation-safety publications alone.
The same aircraft can have different expected loss by operating site, time of day, and route altitude. For example, an inspection flight over an empty industrial yard may expose few uninvolved persons but valuable equipment and infrastructure. A commercial delivery mission near crowds may have lower insured hull value yet a larger human-injury exposure. Accordingly, aircraft replacement value is an incomplete proxy for liability risk, and a mass threshold is an administrative classification rather than a complete loss model.
Human injury, property damage, and indirect economic harm
Bodily injury severity depends on impact kinetics and geometry, medical response, the vulnerability of the injured person, and legal damages rules. Medical expense, future care, lost earnings, non-economic damages, and legal costs may arise from the same occurrence, but they have different documentation and reserving profiles. Property damage may involve direct repair, replacement, investigation and debris removal, and a claim for loss of use. Liability for pure economic loss differs significantly across legal systems. Policy language may exclude or restrict some categories even where they are legally claimed. A research dossier should therefore avoid equating total alleged harm with the insurer's final cost.
Incident accounting is especially sensitive to severity classifications. A serious-injury threshold used to trigger an aviation report is a surveillance rule, not the boundary of civil compensability. Minor injury may be legally actionable and financially material. Conversely, an aviation regulator's finding that the operator did not follow a flight restriction does not by itself determine whether a particular claimed medical condition was caused by the incident or that an exclusion is enforceable.
An insured loss calculation involves the claimant's legally recoverable amount, adjusted for defenses and apportionment, and then the contract's response. Claims-handling and legal defense expenses may be included within or in addition to policy limits depending on the wording. Where many claims arise from one technical malfunction or one swarm event, the distinction between an occurrence limit and a policy-period aggregate becomes decisive. The documentation should identify claimant count, event definition, injury type, geographical location, legal theory, the chronology of claims, and other policies or indemnities potentially responsive.
Who can be responsible?
An operator may incur duties regarding preflight checks, airspace restrictions, maintenance, personnel competence, flight termination, and safe mission design. A remote pilot may have individual responsibilities. The owner may bear responsibility through ownership, contract, or applicable statute. A service provider may furnish route planning, fleet management, communications, cloud navigation, or automated avoidance functions. A manufacturer, software provider, component producer, or maintenance contractor may be implicated when a defect or negligent service contributes to the harm. A client may have operational control or responsibility under the actual contract. Their relationships depend on the jurisdiction and facts; naming every participant in an accident report does not establish liability for each.
The causal analysis should distinguish what failed from why the event was legally attributable to a defendant. Suppose a navigation anomaly makes a drone drift outside its permitted area, while a remote pilot also misses a procedural response. Relevant questions include whether the anomaly was foreseeable, whether the safeguard was available, whether the procedural omission affected the outcome, whether a manufacturer made representations about fail-safe behavior, and whether the operator could reasonably have reduced the consequences. Several causes can coexist. Insurance policies may define insured persons and entities differently from civil-liability law, producing defense disputes even when the underlying facts are substantially agreed.
The distinction also matters for contractual indemnities. A customer may contractually require a drone contractor to indemnify it for a wide range of losses, while the contractor's liability policy may cover only legal liability that would exist independently of the indemnification contract, absent a contractual-liability extension. Product liability can sit under a manufacturer's separate program; an aviation operator's coverage does not necessarily insure design or manufacturing failures as independent liabilities of another entity. Subrogation and contribution determine subsequent allocation of paid losses, subject to policy terms and local law.
Investigated event: Darling Harbour, Sydney
The Australian Transport Safety Bureau's Darling Harbour collision investigation concerns a DJI Inspire 2 engaged in aerial photography on January 15, 2021. During the second flight, the aircraft accelerated unexpectedly and failed to respond to the pilot's control inputs. It ultimately broke a hotel window, caused minor injury to a hotel occupant through flying glass, and was destroyed. The ATSB published its final report on June 23, 2022. This is a documented example of simultaneous hull loss, third-party property damage, and injury arising from a single operating event.
The ATSB attributed the loss of control to compass failure caused by electromagnetic interference. The failed compass also rendered the return-to-home safety function ineffective. The report distinguished the technical contributing factor from other risk-increasing behavior, finding that the pilot did not follow the operator's emergency procedures and did not comply with the conditions of the relevant operating permission. The investigation identified these latter matters as additional safety factors rather than the technical initiating cause of the collision. It expressly treated its findings as accident-prevention evidence rather than an apportionment of civil blame.
From an insurance perspective, the case establishes a sequence for investigation, not an adjudicated policy outcome. The appropriate adjuster would request the actual policy and endorsements, aircraft and payload schedule, insured entity, operator approval, flight logs, preflight and maintenance records, pilot credentials, assignment contract, impact photographs, hotel repair evidence, medical evidence, claims notifications, and any applicable authority correspondence. The record would need to resolve whether the aircraft and activity were insured, what caused the loss, whether an exclusion or condition applied and was legally effective, and whether claimed amounts were causally and legally recoverable. The published ATSB report does not provide an insurer's coverage decision, a settlement amount, or a court judgment. It must not be used to imply one.
This case also explains why a narrow classification of a data-link anomaly can be misleading. The investigation found that the downlink failed or ceased, yet manufacturer analysis indicated that uplink control inputs were still received. A general report that the drone had “lost signal” would obscure important causal details. For cyber, electronics, and autonomous-system risks, the difference between control delivery, command execution, sensor integrity, and safety automation becomes part of claim evidence. A lost telemetry stream can also make subsequent reconstruction harder, creating correlation between event severity and the probability that its cause will be observed.
Collective and cross-risk consequences
Unmanned operations create the possibility of simultaneous or clustered claims. Geographical concentration occurs when many vehicles travel through one urban corridor. Technical concentration occurs when a software update, firmware defect, communications service, or battery design is common to a fleet. Contractual concentration arises when the same insurer covers the operator's drones, third-party liability, cargo, and perhaps the service provider's technology or products. One failure mechanism may affect multiple policies and insureds. The accumulation problem is consequently characterized by shared dependencies as well as by the number of aircraft.
A fleet's expected average loss can appear stable while its extreme quantiles change substantially. A launch schedule that sends ten aircraft through the same area in adverse weather has different dependence from ten flights distributed over separate locations and times. Underwriting therefore needs to record fleet size, simultaneous missions, route overlap, weather constraints, common vendors, recall or grounding exposure, and whether a claims “occurrence” aggregates multiple damaged subjects. The exact application of aggregate limits, reinsurance attachments, and exclusions remains wording-dependent.
Coverage Triggers and Insurance Contract Architecture
Hull, liability, and related interests
A first-party hull contract may indemnify direct loss of or physical damage to the specified aircraft, including covered ground and flight events, subject to agreed valuation, excess, conditions, and exclusions. Coverage may extend to a specified controller, batteries, or payload only when included within definitions and schedules. A liability contract concerns covered legal obligations to compensate others for bodily injury or property damage resulting from insured operations, potentially alongside defense and supplementary payments. These grants can appear in one policy but retain separate insuring conditions and limits. A physical loss to the operator's drone does not, by itself, establish liability to a third person; nor does a third-party claim require complete destruction of the hull.
The market's standardized drafting provides a particularly useful documentary contrast. The International Underwriting Association lists the 2024 AVN146 commercial UAS policy and its optional endorsements. In 2025 the Aviation Insurance Clauses Group released AVN157 for unmanned aircraft, with the IUA's publication circular identifying liability and physical damage as core elements. These primary publication records show available model clauses and their subject matter. They do not verify execution of particular endorsements or permit an analyst to assert a coverage grant beyond the inspected document and actual insurance contract.
The endorsements reveal important divisions in insured exposure. An unauthorized-use or theft extension concerns events in which the lawful insured operator may have lost possession or control of the aircraft. A regulatory-breach endorsement concerns how breaches of aviation requirements affect insurance, subject to its negotiated text and governing law. A medical-payments or supplementary-payments extension can alter the handling of claims that do not fall neatly within the main liability grant. An extended aviation liability endorsement may enlarge the protected liability class. A cargo-liability endorsement addresses transported property separately from the drone's own hull. The product taxonomy itself is evidence that a single generic label, “drone insurance,” does not determine the breadth of cover.
A disciplined coverage inquiry
Claims analysis begins with the insuring agreement. What was the scheduled aircraft, the covered activity, and the relevant time period? Which insured person or legal entity operated the drone? Where did the incident occur, and were territorial conditions met? Did an accident, occurrence, loss, claim, or notification trigger apply? Which losses fall within bodily injury, property damage, physical hull damage, or supplementary cover? The answer then proceeds through exclusions, limits, retentions, conditions, and applicable legal rules for interpretation. The policy's actual wording, not the insurer's marketing description or certificate alone, controls this investigation.
The distinction between an occurrence policy and claims-made form may matter where a technology or professional liability component joins aviation liability. Aviation casualty cover often centers on the occurrence of bodily injury or property damage during the policy period, whereas a separate professional, cyber, or technology policy may require a claim to be made and reported within defined periods. An insurer must verify the actual coverage trigger for the policy under review. The creation of a new commercial service, or an increasing degree of autonomy, does not justify inferring a universal trigger across all lines.
A practical analytical sequence is to define the event; identify all potential insureds; establish each alleged injury and damage; identify the governing jurisdiction and legal cause of action; identify the applicable policy periods and declared operations; test insuring agreements and exclusions independently; apply deductibles and limits; examine defense treatment and other insurance; and separately calculate likely gross recovery, insurer indemnity, and retained loss. This order minimizes errors such as treating insured hull damage as proof of a covered liability obligation or using a regulatory fine as a proxy for compensatory damages.
Compliance clauses and the causal significance of breach
Drone operations are intensely regulated, which creates a frequent temptation to treat any noncompliance as automatic exclusion from all insurance. Contractual interpretation requires more precision. A policy may make compliance an express condition, define authorized operations as part of the insuring agreement, exclude specific illegal flights, or contain an endorsement modifying the consequence of a breach. Jurisdictional rules governing warranties, conditions, materiality, causation, and remedies can also affect the outcome. Consequently, the effect of a regulatory breach must be established by the executed contract and applicable law. The existence of an AVN159 regulatory-breach endorsement and the earlier AVN149 counterpart demonstrates that this question receives specialized contractual treatment.
The Darling Harbour case is analytically revealing. ATSB attributed the technical loss of control to sensor interference but also recorded breaches of operating limitations. Insurers and claimants would have to consider the specific wording, whether and how each breach affected the insured event, and the legal force of contractual conditions. The public safety finding alone neither grants nor denies insurance. This separation protects the integrity of claims analysis and permits more coherent claims reserving when technical and coverage disputes develop on different timelines.
Third-party damage and the boundary of other policies
A drone can damage property that is temporarily entrusted to an operator, carried as payload, or owned by a customer. Liability for property in the insured's care, custody, or control may be treated differently from liability for an unrelated building or passerby. Cargo transportation, delivery delay, and resulting contractual penalties can call for separate products or extensions. Privacy invasion, data gathering, and cyber compromise are relevant risks of camera and connected drones, but a standard bodily-injury and property-damage insuring agreement cannot automatically be read as broad privacy or cyber insurance. Drone operations may also produce regulatory investigations or administrative penalties, whose insurability and coverage depend on law and wording.
For a manufacturing defect, the operator's liability insurance and the manufacturer's product liability contract may interact through defense, indemnity, contribution, or subrogation. Shared failures can create overlapping claims across aircraft hull, aviation liability, product liability, and errors-and-omissions lines. An insurer seeking to model aggregate exposure should identify common causal mechanisms and all potentially implicated policies rather than assume that different insurance classifications guarantee independent loss experience.
Limits, deductibles, and adequacy of security
A compulsory statutory minimum serves a regulatory function. It is not an actuarially tailored guarantee that every extreme loss is fully financed. Under Article 7 of Regulation (EC) 785/2004, the relevant minimum is expressed in SDR rather than a fixed dollar or euro amount. Its converted purchasing power changes with the SDR exchange rate; the actual amount and application must be established for the operation in question. Aggregate coverage, excluded perils, deductibles, multiple claimants, and legal expense treatment are separate from an advertised “limit.”
A claim involving multiple people and extensive property damage may exhaust the per-occurrence limit before all legal claims are resolved. Conversely, an aircraft with a very small replacement cost can be associated with liability limits much larger than its value. The adequacy of the limit should therefore be evaluated using credible severity scenarios, including severe bodily injury, damage to critical infrastructure, collision with another aircraft, and simultaneous claims. Reinsurance and insurer capital resources affect the financial performance of insured obligations but do not expand an individual contract's legal coverage by themselves.
Evidence Appraisal, Actuarial Analysis, and Sensitivity
The meaning of a measured accident rate
The empirical object of an insurance rating exercise is covered monetary loss relative to an explicitly defined insured exposure. A civil-aviation accident registry usually observes a different object: accidents satisfying a reporting rule. A scientific ground-risk simulation may observe the probability of an impact, the spatial distribution of crash locations, a modeled injury probability, or a safety threshold. These measurements are related, but conversion among them requires observable or assumed intermediate quantities. A reporting threshold also creates left truncation: minor impacts and near misses may be absent even when insurers receive related demands or incur defense expense.
In the scientific literature, la Cour-Harbo's quantitative analysis of ground casualties treats risk through aircraft failures, descent/impact, exposed population, and consequences. Jiang, Blom, and Rattanagraikanakorn's impact-area and fatality assessment highlights how design, impact footprint, and conditional human vulnerability interact with routing. Liu and coauthors' uncertain horizontal-distance estimation models spatial uncertainty in ground risk; Zhang and coauthors' national-airspace safety assessment develops a complementary quantitative perspective on unmanned aircraft in shared airspace. These contributions establish mechanisms and modeling approaches, rather than a ready-made table of claim frequencies or insured loss severities. One cannot directly use an expected casualty probability as an expected third-party liability payment without a legal, contractual, and monetary mapping.
A useful decomposition, employing conventional frequency–severity reasoning, is to define N as the number of missions or flight hours insured during the observation interval, q as the probability of a loss-of-control or relevant hazardous event per exposure unit, r as the probability that such an event reaches third-party persons or property, and s as the probability that it causes a compensable injury or property loss conditional on contact. Under the deliberately restrictive assumption that these probabilities are stable and conditional factors have been consistently specified, the expected number of potentially compensable events is Nqrs. Let Z be the total legally recoverable third-party amount for such an event, including the legally allowable injury and property components, and let I(Z;W) be the insurer's actual indemnity under wording W after exclusions, deductibles, sublimits, and limits. Then an illustrative expectation is
This expression is an accounting decomposition of conditional expectation under its stated simplifications. It is not a new model, an empirically fitted rating formula, or a statement that q, r, and s are independently identifiable in practice. A more general representation integrates a conditional event rate and a conditional insured severity over route, time, operating environment, aircraft, and operator characteristics. The distribution of Z may include zero-valued claims, disputes about liability, and positive defense cost when indemnity is ultimately zero. Written premium must also provide for defense, claim adjustment, operational expenses, catastrophe cost, reinsurance, uncertainty, and capital. Treating Nqrs as a complete premium ignores those components.
Population footprints and spatial selection
One physical approximation for exposed ground population uses the crash impact area A (square meters) and local human density rho (persons per square meter). Under a locally homogeneous distribution and a random impact center independent of occupant locations, the expected number of persons within the footprint is rho A. If the probability of bodily injury to an individual inside the affected footprint is p, expected injuries per hazardous crash can be approximated by rho A p. This is a local first-moment relation; its interpretability depends on assumptions about occupancy, shielding, vertical structures, time of day, crash orientation, vehicle energy, and the probability that the aircraft's trajectory intersects the populated cell. It should not be extrapolated to fatalities without a separately justified conditional fatality mechanism.
For illustration, a 12-square-meter effective impact footprint over a cell containing 0.02 persons per square meter yields 0.24 expected persons in the footprint under the homogeneous assumption. If the conditional injury probability in this illustrative situation is 0.10, the expected number of injuries per relevant impact is 0.024. The units cancel as follows: square meters times persons per square meter times injuries per exposed person produces expected injuries. Neither 12, 0.02, nor 0.10 is presented as an observed value for any drone category. The exercise shows why a crash frequency recorded across sparsely populated routes cannot be transferred unchanged to an urban delivery network.
A more detailed route calculation partitions exposure into cells indexed by j, with time t_j, failure intensity lambda_j per flight hour, ground-contact probability c_j, and exposure-consequence factor h_j. Expected events are approximated by the sum of t_j lambda_j c_j h_j. A cell may have small time exposure and high third-party hazard; averaging the entire route's population density can conceal this concentration. The EASA SORA 2.5 source framework is relevant as an operational method to organize such ground and air considerations, but its qualitative assurance requirements and operational risk categories should not be treated as actuarially observed insurer payment rates.
A fully disclosed frequency–severity calculation
Consider a hypothetical commercial portfolio operating 100,000 flight hours in one year. Assume 0.0004 potentially compensable third-party events per hour, so the expected claim-event count is 40. Suppose the hypothetical severity distribution, conditional on such an event, assigns probabilities 0.90, 0.09, and 0.01 to legal damages of USD 5,000, USD 100,000, and USD 2,000,000, respectively. The expected gross legally recoverable damages per event are USD 33,500: 0.90 times 5,000 plus 0.09 times 100,000 plus 0.01 times 2,000,000. The portfolio's expected gross damages are USD 1,340,000 before any expense load, discounting, or insurance contract modification. All values are hypothetical and deliberately simple. They demonstrate algebra, not calibration.
Now assume an occurrence indemnity contract with a USD 10,000 deductible borne by the insured and a USD 1,000,000 maximum indemnity for each occurrence; for this example, ignore legal costs and assume no exclusions. Define the insurer payment as I(Z)=min(max(Z–10,000,0),1,000,000). The three corresponding indemnities are USD 0, USD 90,000, and USD 1,000,000. The expected indemnity per covered event becomes USD 18,100: 0.90 times 0 plus 0.09 times 90,000 plus 0.01 times 1,000,000. Expected annual indemnity is USD 724,000. The gap between gross damages and insurer payments reflects both retention and truncation by the policy limit. Changing the severity mix, insured legal jurisdiction, or defense treatment changes the result. A fixed minimum insurance limit should never be substituted for a severity distribution.
For an actuarial sensitivity check, double event frequency while holding all conditional severities and cover conditions constant. Expected indemnity doubles from USD 724,000 to USD 1,448,000. Double instead the probability of the USD 2,000,000 outcome from 1% to 2%, reducing the 90% low-severity class to 89% and holding the 9% middle class constant. Expected insurer indemnity rises to USD 1,124,000. Each additional percentage point of high-severity probability contributes USD 10,000 to conditional indemnity; multiplied by 40 expected events, the increase is USD 400,000. This calculation illustrates the sensitivity of expected policy payments to the frequency of very severe losses. These deterministic sensitivities omit the variability of event count and severity and cannot determine a capital requirement.
For completeness, the corresponding gross expected damages with the adjusted mix are USD 53,450 per event: 0.89 times 5,000 plus 0.09 times 100,000 plus 0.02 times 2,000,000. Across 40 claims, that gives USD 2,138,000. This comparison shows the asymmetry created by policy limits: gross damage expectation increases by USD 798,000, while the insurer's capped expectation increases by USD 400,000. The risk transferred to the insurer and the loss imposed on injured third parties differ in both scale and tail behavior. Accordingly, high policy-limit layers and claimant recovery merit examination even when a low hull value gives an appearance of moderate risk.
Dependence, accumulation, and tail uncertainty
If individual drone losses were independent and identically distributed, aggregate claim counts might be approximated by a Poisson distribution and severity by an independent mixture or continuous distribution. This benchmark is useful for pedagogy but poorly suited to a fleet with shared weather, firmware, navigation software, communication infrastructure, maintenance procedures, charging systems, or operator scheduling. A single common defect may cause multiple loss events or simultaneous liabilities. Parameter uncertainty can itself generate overdispersion even when events are conditionally independent given an unobserved fleet risk level. A model that assumes independence may therefore materially understate an aggregate upper-tail quantile.
A practical accumulation analysis distinguishes common-event correlation (one disruption affecting many aircraft), common-cause correlation (several events produced by a defect), and insurer-side financial correlation (several policy classes affected by one event). Their impacts on coverage depend on the definition of an occurrence, limits per aircraft or event, annual aggregates, exclusions, and the structure of any reinsurance. Reinsurance recoveries need counterparty, attachment, aggregation, and settlement assumptions in addition to expected gross event damage. These parameters are not inferable from a SORA classification alone.
Portfolio segmentation should include aircraft mass and kinetic energy, propeller and battery characteristics, parachutes and flight termination, payload, planned altitudes, route geometry, density and characteristics of persons overflown, night operations, VLOS/BVLOS status, operation category, remote-pilot expertise, autonomy level, fleet software version, and history of safety interventions. Underwriting also needs the exact legal entity and claims jurisdiction. If data are scarce, hierarchical or credibility methods can borrow strength across segments only with transparent similarity judgments and an explicit allowance for unobserved heterogeneity. A model trained on experienced operators in rural work may fail when applied to a new urban operator, even if the aircraft model is identical.
Adverse selection, moral hazard, and the role of observable controls
Insurance demand may be correlated with privately known operational risk. An operator intending dense urban missions, experimental payloads, or aggressive scheduling may value third-party cover more than an occasional rural operator. Insurers mitigate adverse selection by defining intended operations, collecting route and activity information, examining authorizations, and differentiating terms. After cover starts, incentives also matter: deductibles, documented maintenance, pilot competence, telemetry controls, and incident-reporting duties may encourage continuing risk management. These mechanisms involve economic reasoning and cannot establish misconduct by a particular policyholder. A policy's coverage depends on legally effective language, not a broad presumption about incentives.
Telemetry is most useful when it captures actual exposure rather than merely evidence favorable to the insured. A valid denominator must include unsuccessful flights, near misses, aborted missions, and periods of degraded operation, within the boundaries agreed for the exposure measure. Claims data alone may contain no reliable information about safely completed missions. Conversely, telemetry from successful missions without matched loss observations cannot estimate liability severity. Missing-event data may be systematically related to the severity of crashes, as the ATSB's Darling Harbour evidence demonstrates through distinctions between control link directions and recorded onboard data.
Event evidence and the burden of reconstruction
An analytically complete claims file should preserve a time-indexed chain from operational intent through commands and aircraft-state logs, sensor status, geolocation accuracy, image and video records, equipment identity, site conditions, impact, medical or property evidence, and subsequent repairs or payments. Data provenance should identify clock synchronization, software and firmware version, export method, custody transfers, missing segments, and any post-event alterations. The log is one item of evidence. It is interpreted alongside independent testimony, physical evidence, safety investigation findings, and governing legal tests.
Safety regulators may publish findings that remove doubt about a technical mechanism while leaving civil liability unresolved. For the Sydney collision, the ATSB's report presents a sufficiently concrete causal account to demonstrate an insured event scenario, but does not supply the insurance wording, policy limits, insured schedule, claimant demand, or adjudicated liability. A quantitative claim severity drawn from the incident would consequently be fabricated. The example supports the necessity of a dual track: physical cause assessment and legal-contractual adjudication, with claim reserving updated when evidence resolves either track.
The credibility of sources is ordered by the question they answer. Statutory and regulatory texts establish legal duties within their jurisdiction and effective period. Official regulator guidance explains administrative interpretation and operational practice but may simplify legal exceptions. Executed policy terms would establish the coverage contract, whereas published model clauses establish possible drafting. Official accident investigations provide causal evidence within their expressly limited remit. Peer-reviewed articles provide mechanisms and models whose assumptions require scrutiny; they do not provide universal market-level loss estimates. This hierarchy is task-relative rather than a single ranking of the overall merit of institutions.
Claims reserving and basis of observation
A third-party liability claim can take longer to settle than the underlying hull loss because bodily injury develops, ownership and responsibility are contested, or several defendants exchange contribution claims. Case reserves should record the best available liability, causation, coverage, damages, defense, and recovery assumptions by claim, with an explicit uncertainty range. Incurred-but-not-reported liabilities require observation of reporting delays, including claims that initially appear as minor property damage but later involve injury or consequential loss. Portfolio-level loss triangles from mature aviation insurers may be difficult to transfer to drone portfolios without adjustment for developing technology, new operating permissions, changes in reporting requirements, and evolving policy terms.
An underwriter observing a low paid-claims rate cannot conclude that technical safety is high if the policy excludes important perils or claims are delayed. An aviation regulator observing a low serious-incident rate cannot conclude insurance claims will be small if minor property damage is frequent or legal costs are substantial. Conversely, an insurer's high claim count may partly reflect superior reporting, broader coverage, or a lower deductible. Sound comparison normalizes exposure, event definition, coverage terms, claim development, and reporting incentives.
Sensitivity and identification limits
The paper's numerical calculation is fully reproducible from its declared assumptions, but its input assumptions are not estimated from source data. The illustrative 0.0004 event rate, severity probabilities, damages, deductible, limit, footprint, density, and conditional injury probability are fictional. They are not estimates of actual drone safety, required insurance, or an actuarially indicated premium. The expected values describe a simplified portfolio in one period. Inflation, jurisdictional differences in damages, legal defense, claim settlement delay, social inflation, catastrophic aviation collision, environmental damage, and insurer insolvency could alter an actual risk assessment. Catastrophic tails in particular demand scenario reasoning when data are sparse.
Published research gives technical plausibility to the relevant conditional mechanisms, especially impact geometry and population exposure. It does not resolve the empirical identification problem of estimating insurance claim frequency, legally recoverable severity, and policy indemnity from the same representative cohort. A future study would require a privacy-protecting linked dataset containing exposure hours and route cells, failure/accident reports, claims and payment amounts, policy wordings, and consistent legal outcome codes. Any claim of measured actuarial accuracy should be postponed until such data and out-of-sample evaluation exist.
Scientific Synthesis and Research Agenda
What the comparative evidence establishes
The comparison supports five connected conclusions. First, the eligible insured operation is defined by a dated legal category, the real operator, the mission, and actual policy conditions. European Union, United Kingdom, United States, Canada, and Australia show material differences in the relation between aviation authorization and insurance duty; a generic assertion of universal compulsory cover cannot survive the primary-source comparison. Second, third-party harm depends on physical event mechanisms and the distribution of people and property across geography. The value of a vehicle is a poor stand-alone proxy for legal-liability severity. Third, liability attaches to legally responsible entities through jurisdiction-specific causes of action and evidence, whereas insurer indemnity depends on contract wording and limits. Fourth, autonomy and shared digital infrastructure shift the mix of foreseeable failure mechanisms and create accumulation pathways. Fifth, empirical knowledge of insured loss remains constrained by inconsistent exposure denominators and the incomplete intersection of safety and claims data.
The research identifies a series of transitions that should be documented in every scientific account of an incident: hazardous operation, failure or collision, third-party damage, legally attributable liability, insurable liability, policy response, amount payable, and recovery from other parties. Each transition uses its own evidence, is subject to its own uncertainties, and may fail independently of the others. This is an organizational description of established questions in risk analysis, insurance law, and claims practice, not a new theoretical framework or doctrine. Its importance is practical and methodological: it prevents evidence collected for one purpose from being silently repurposed as proof of another.
The interplay of prevention and indemnification
Risk mitigation lowers the frequency or consequence of some hazards. Geographical routing can reduce exposure to populations; design changes may reduce impact energy or probability of injury; crew training and emergency procedures may lower the risk of uncontrolled outcomes; and safety assurance can make operations more predictable. Yet prevention and insurance deal with different residual uncertainties. A risk assessment may find an operation acceptable under a stated safety method, while an insurer must still price and finance its residual legal liabilities. Conversely, a high insurance limit does not authorize a flight in prohibited airspace. This dual responsibility becomes particularly salient as commercial flights scale beyond individual supervised missions.
Regulators, researchers, operators, and insurers can improve mutual understanding by agreeing on event units and transparent measurement standards. In principle, flight exposure should be classified by aircraft characteristics, geographic ground exposure, mission phase, operating permissions, software versions, and causal incident mechanism, while claim observations separately record legal responsibility, defense costs, indemnity paid and reserved, coverage exclusions, and recoveries. This suggested research design follows from the observed differences between safety records and insurance datasets. It is a data-quality recommendation, with no claim that a standardized international dataset already exists.
Comparative legal and contractual boundaries
Europe combines supranational operational rules with insurance regulations and material national variation. The United Kingdom's commercial-use focus, the FAA's primarily operational Part 107 framework, Transport Canada's recommendation, and CASA's employer/client verification guidance produce different entry points for insurance evidence. The actual insurance outcome in every system remains sensitive to the claimant's legal entitlement, the insurer's contract, and applicable mandatory insurance or claims rules. Political and operational cross-border variation means that an internationally active operator may require more than an extension of territorial coverage: it may need a verified local duty, permitted aviation activity, and a contract enforceable against the relevant liability.
The International Underwriting Association's separate 2024 and 2025 model policies and optional endorsements provide a documentary basis for studying how the industry separates hull, liability, extra expense, regulatory breach, and nonstandard operations. The clauses also make clear that scientific commentary should never imply an endorsement was incorporated into a particular policy solely because it exists in a model library. Primary-source comparison of the published clause materials is possible; claims about insurer-specific adoption, premium, claims settlement, or the actuarial impact of a clause require additional empirical evidence.
Scientific priorities
Several focused research questions follow. Which combinations of route-density exposure and aircraft design most strongly predict third-party injury claims under observed operating conditions? How do claims and safety data differ when their exposure denominator, inclusion threshold, and reporting incentives are made consistent? What fraction of severe loss arises from recurrent equipment faults compared with operator error, weather, cyber dependency, and infrastructure interruption? Under what contractual language are common-cause fleet incidents aggregated into one occurrence? How much of the total economic cost of serious drone accidents is retained by operators or injured parties because of deductibles, exclusions, inadequate limits, or gaps between compulsory and actual cover? How does the move from remote piloting toward higher autonomy change the practical allocation of liability among operators, manufacturers, software suppliers, and fleet supervisors?
These questions require primary accident records, exposure-linked policy and claim histories, and reliable legal outcomes. Confidential claim files and many commercial models are not presently available within this study. Hence no statement of measured global drone-claim frequency, insurance penetration, premium adequacy, or loss-ratio trend is made. The absence of these estimates is a substantive finding about the present evidence boundary, rather than an invitation to substitute promotional market figures.
Original analytical contribution and boundaries
The original contribution of this monograph is a source-grounded integration of operational authorization, physical ground/air exposure, insured interests, civil-liability attribution, model policy drafting, claims evidence, and actuarial loss measurement in one comparative account. The analysis makes concrete the non-equivalence of safety incident counts, human harm, civil-liability claims, and insurance indemnity, with an audited numerical illustration of the economic effects of a retention and a limit. It uses the Sydney occurrence as a bounded case for explaining mixed hull/third-party harm, sensor causation, and insurance-evidence requirements. It does not propose a new risk doctrine, proprietary model, conceptual system, empirical discovery, or scientific priority claim.
Conclusion
Unmanned aircraft insurance covers a developing field in which a relatively modest movable asset may generate legal obligations considerably larger than its replacement cost. The scientific subject includes the aircraft and its systems, the insured persons and interests, the people and property at risk, the legal rules distributing responsibility, and the contractual terms by which defined liabilities become insurer payments. Treating these components as independently observable prevents unjustified transitions from regulatory compliance to coverage, from an accident to a compensable claim, or from an approved safety case to an assumed premium.
The authoritative evidence establishes real differences among national insurance requirements, identifies developing industry model clauses for commercial and consumer unmanned aircraft, and demonstrates through an officially investigated collision how technical failure can simultaneously destroy a drone, injure a person, and damage a third party's property. Peer-reviewed ground-risk analyses describe relevant physical determinants, including crash footprint, route density, and conditional harm, while stopping short of providing insurance claim distributions. The paper's transparent frequency–severity example further shows how retention and limited indemnity redistribute loss between injured parties, operators, and insurers. Future work should link exposure, aircraft and software characteristics, incidents, policy language, legal outcomes, and claim payments with stable definitions and verified provenance. That empirical foundation would strengthen both underwriting and public understanding of the financial consequences of civil unmanned aviation.
References
Australian Transport Safety Bureau. (2022, June 23). Loss of control and collision with terrain involving DJI Inspire 2, remotely piloted aircraft, Darling Harbour, Sydney, New South Wales, on 15 January 2021 (Investigation AO-2021-001). Australian Government. https://www.atsb.gov.au/investigations/ao-2021-001
Civil Aviation Authority (United Kingdom). (2026). Insurance requirements. UK Civil Aviation Authority. https://www.caa.co.uk/drones/open-category/moving-on-to-more-advanced-flying/insurance-requirements/
Civil Aviation Authority (United Kingdom). (2026). Flying drones for work. UK Civil Aviation Authority. https://www.caa.co.uk/drones/open-category/moving-on-to-more-advanced-flying/flying-drones-for-work/
Civil Aviation Safety Authority (Australia). (2026). Hiring a drone operator. CASA. https://www.casa.gov.au/operations-safety-and-travel/consumer-and-passenger-advice/hiring-drone-operator
European Commission. (2004; consolidated 2020). Regulation (EC) No 785/2004 on insurance requirements for air carriers and aircraft operators. Official Journal of the European Union. https://eur-lex.europa.eu/legal-content/en/TXT/?uri=CELEX%3A02004R0785-20200730
European Commission. (2019, May 24; subsequently amended). Commission Implementing Regulation (EU) 2019/947 on the rules and procedures for the operation of unmanned aircraft. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg_impl/2019/947/
European Union Aviation Safety Agency. (2025). ED Decision 2025/018/R: Amendment of the SORA methodology and associated acceptable means of compliance and guidance material. EASA. https://www.easa.europa.eu/en/document-library/agency-decisions/ed-decision-2025018r
European Union Aviation Safety Agency. (2026). Easy Access Rules for Unmanned Aircraft Systems: Specific Operations Risk Assessment (SORA). EASA. https://www.easa.europa.eu/en/document-library/easy-access-rules/online-publications/easy-access-rules-unmanned-aircraft-systems?erules-id=ERULES-1963177438-24086
European Union Aviation Safety Agency. (2026). Insurance: What should you consider when getting drone insurance? EASA. https://www.easa.europa.eu/en/domains/civil-drones/privacy/insurance
Federal Aviation Administration. (2026). Small Unmanned Aircraft Systems (UAS) regulations (Part 107). FAA. https://www.faa.gov/newsroom/small-unmanned-aircraft-systems-uas-regulations-part-107
Federal Aviation Administration. (2026). When do I need to report an accident? FAA. https://www.faa.gov/faq/when-do-i-need-report-accident
Federal Aviation Administration. (2026). DiSCVR: Drone identification system for first responders and law enforcement. FAA. https://www.faa.gov/uas/public_safety_gov/discvr_tool
International Civil Aviation Organization. (2015). Manual on Remotely Piloted Aircraft Systems (RPAS), Doc 10019 (1st ed.). ICAO. https://store.icao.int/en/manual-on-remotely-piloted-aircraft-systems-rpas-doc-10019
International Civil Aviation Organization. (2024). Annex 13 to the Convention on International Civil Aviation: Aircraft Accident and Incident Investigation. ICAO. https://store.icao.int/en/annex-13-aircraft-accident-and-incident-investigation
International Underwriting Association. (2024, October 1). AVN146 Unmanned Aircraft Insurance Policy and related commercial-operation clauses. Aviation Insurance Clauses Group. https://www.iua.co.uk/IUA_Member/IUANew/Clauses_Documents/IUA-Clauses-Showcase.aspx?id=2&ordinal=723
International Underwriting Association. (2025, September 29). AVN157 Unmanned Aircraft Insurance Policy. Aviation Insurance Clauses Group. https://www.iua.co.uk/IUA_Member/IUANew/Clauses_Documents/IUA-Clauses-Showcase.aspx?id=6&ordinal=1080
International Underwriting Association. (2025, October 1). Model Unmanned Aircraft Insurance Policy (Circular 079/25). https://www.iua.co.uk/IUA_Member/IUANew/NewsPages/Circulars_2025/Model-Unmanned-Aircraft-Insurance-Policy.aspx
Jiang, C., Blom, H., & Rattanagraikanakorn, B. (2025). Enhancing safety feedback to the design of small, unmanned aircraft by joint assessment of impact area and human fatality. Risk Analysis, 45(5), 1115–1131. First published online September 14, 2024. https://doi.org/10.1111/risa.17649
la Cour-Harbo, A. (2019). Quantifying risk of ground impact fatalities for small unmanned aircraft. Journal of Intelligent & Robotic Systems, 93, 367–384. https://link.springer.com/article/10.1007/s10846-018-0853-1
Liu, Y., Zhang, X., Wang, Z., Gao, Z., & Liu, C. (2021). Ground risk assessment of UAV operations based on horizontal distance estimation under uncertain conditions. Mathematical Problems in Engineering, 2021, Article 3384870. https://doi.org/10.1155/2021/3384870
Transport Canada. (2026). Tips and best practices for drone pilots. Government of Canada. https://tc.canada.ca/en/aviation/drone-safety/learn-rules-you-fly-your-drone/tips-best-practices-drone-pilots
Zhang, X., Liu, Y., Zhang, Y., Guan, X., Delahaye, D., & Tang, L. (2018). Safety assessment and risk estimation for unmanned aerial vehicles operating in national airspace system. Journal of Advanced Transportation, 2018, Article 4731585. https://doi.org/10.1155/2018/4731585