An agent-based model for technology adoption in Air Traffic Management

Technology adoption depends not only on the development of new technologies, but also on the existence of regulations able to foster the implementation of such technologies. To facilitate the exploration of different policy options aimed at accelerating the adoption of new Air Traffic Management (ATM) technologies, an agent-based model that represents the behaviour of the European ATM system has been developed. This model includes representations of the main stakeholders in the ATM ecosystem: regulatory bodies, technology providers, labour unions and technology adopters, including Air Navigation Service Providers (ANSPs), airlines and airports. New ATM technologies, policies imposed, behavioural biases (e.g., loss aversion) and exogenous variables (e.g., fuel price) drive the actions of the agents, leading to the emergent global behaviour of the system. A calibration and validation process involving historical data, gaming experiments and participatory simulations was performed. The model was used to evaluate various policies that included economic incentives and penalties in two scenarios: one based on past events and another focused on the future. The results allow us to analyse which individual stakeholders benefit the most from each policy and to identify the mechanisms that emerge and drive the path of technology adoption, finding that a combination of economic incentives and penalties provides promising economical and operational results.


Background and motivation
In recent years, the need to accelerate ATM technological change has become more and more evident: growing traffic demand and new market entrants, such as commercial drone applications, are rapidly taking the ATM system to its limits, calling for disruptive solutions able to boost the performance of ATM operations.Emerging technologies, especially digitalisation and automation, have the potential to facilitate this urgently needed technological upgrade.However, technology evolution is a necessary but not sufficient condition: innovation is a complex phenomenon, which depends not only on the development of new technologies, but also on the existence of regulation and institutions able to facilitate and foster the implementation of such technologies.
The need for both regulatory and technological developments that lead to the modernisation of the ATM system has been recognised since long by the European Commission, and is at the origin of the coordinated launch of the Single European Sky initiative and its technological pillar, SESAR.However, despite all the efforts undertaken within both initiatives, the results have not lived up to the expectations.The European Commission [1] recognised the need to facilitate the transition from the SESAR development phase to deployment.In this context, the development of an in-depth understanding of the factors that drive technology adoption in ATM and the identification of mechanisms to accelerate the R&I lifecycle emerge as critical needs.
When elaborating their measures, policy makers have to deal with a complex system.Society is a network with many political, social and economic interactions and any new regulation may create unexpected changes in the network.To assist them in their task, agent-based models (ABM) offer a number of features that make them particularly suitable for the study of innovation uptake processes, see [2].The possibility to model agents' heterogeneity, non-rational behaviours or biases (e.g., loss aversion), learning processes, evolutionary behaviour and path dependence are some examples.
The aim of this paper is to present a policy assessment framework based on ABM to compare different policy measures designed to incentivise technology adoption.The goal of such framework is not to increase adoption by itself, but to compare the disaggregated distributional effects and the aggregated social welfare brought about by the different policies under study.Although ABM has been already used for studying technology adoption and policy assessment in networked sectors, this is the first time that this technique is applied to the ATM field.

Agent-based model
The policy assessment model proposed aims at supporting the design of measures and regulations intended to accelerate the adoption and deployment of new ATM technologies.The model enables the evaluation of the effects of a variety of policies (e.g., innovative ATM pricing schemes) on the level of adoption of the technologies under study and the resulting impact for ATM stakeholders in terms of economic benefits and costs.A high-level scheme of the proposed model architecture is shown in Figure 1.
The proposed model is based on the following assumptions and constraints: • It is focused on the adoption of technology by civil aviation.
• Global trends (e.g., economic growth) will not be affected by the evolution of ATM, considering its relative small size compared to the global aviation industry.• Major disruptions in aircraft design are not considered, given its uncertainty and long-term plan.Evolution in fuel efficiency is taken into account, based on the forecasts provided by the main aircraft and engine manufacturers.• National interests are not considered.
• Airlines choose the route that minimises their cost function (fuel + navigation charges [3]).

Scenario definition
The scenarios to be simulated by the model are defined by: (i) the policies to be tested, (ii) the available technologies, and (iii) the evolution of the exogenous variables.
The policies to benchmark have been selected based on their relevance, following a combination of literature review and stakeholder consultation [4] and include: (i) Flexible charging, i.e., ANSPs are allowed to add a certain margin to their unit rates if they adopt a given technology; (ii) Best equippedbest served, i.e., the charges for different airspace users would be asymmetric depending on their ATM equipment; (iii) Subsidies, i.e., the stakeholders would receive financial aid subject to ATM technology adoption.;and (iv) Increased involvement of certification authorities in the research phase: the expected effect of this measure is modelled as the earlier availability of certain ATM solutions and reduced risk perception on the adopters' side.
Each technology has been modelled considering the attributes that affect agents' decision towards adoption or rejection, i.e., costs (aircraft retrofit, airport implementation cost, ACC implementation cost and training); benefits (aircraft fuel savings, delay reduction, airport capacity increase, airport cost reduction and ATCO hour productivity increase); compatibility with existing technology; in service life (to calculate amortisation costs); implementation requirements (time required to the entry into service of the new technology, typically ranging from some months to a few years); and effect on labour conditions (the higher the conflicts with the workers the lower the acceptance by the public, then hindering adoption).
The exogenous variables are the external conditions that affect agents on their behaviour.In the model, we have considered: • Passenger demand.It is mainly related to the Gross Domestic Product (GDP) and population changes of the regions.The effect of the COVID-19 pandemic has not been considered.

• Engine efficiency. Engines fuel consumption is retrieved from the ICAO Aircraft Engine
Emissions Databank.This data, with the annual reduction rate of average fuel burn of new aircraft estimated by [5], allows us to estimate past and future engine efficiencies.• Fuel price.The historical and forecasted Jet-A fuel prices (€/kg) up to 2050 provided by the US Energy Information Administration (EIA) are considered.• Unitary labour costs.The labour costs of different positions are needed for the cost calculation of the agents: pilot unitary costs, non-cabin staff unitary costs, ATCO unitary costs, non-ATCO ANSP staff unitary costs, and airport staff costs.

Agents
The agents included in the model can be classified into four different groups, according to their role in the technology adoption problem: adopters, industry (technology providers), labour unions, and regulator.The main criteria followed for the definition and classification of the agents included in the model are as follows: • The adopters' group is formed by the stakeholders in a position to adopt new ATM technologies, which includes airports, airlines, and ANSPs.• En-route and terminal ANSPs are split, due to the operational differences and the liberalisation of ATC services in some European countries.Interviews with ATM stakeholders [6] showed that the technology adoption dynamics may differ depending on the ownership model (private vs public) and the revenue scheme (cost recovery vs liberalised market).

Airports
The ownership of European airports has changed drastically in the last decade, rising the number of passengers that make use of Public-Private Partnerships (PPPs) or fully private airports up to a 75% of the total passengers carried in 2016 [7].The privatisation of the sector makes the airports dependant on investors and, thus, it is reasonable to assume that they behave as profit maximisers.This goal drives their decisions in the model: • Establishment of airport fees.The fees are set to maximise profit projections.The cost structure of the airport is considered (labour costs, non-staff operational costs and depreciation cost) as well as its aeronautical revenues (dependant on the charges) and non-aeronautical revenues, which are considered as proportional to the number of passengers carried.• Adoption of new technology.New technologies are considered to improve airport's capacity and reduce its operational costs.On the other hand, technologies have to be amortised over their in-service life.Other aspects of the technology are translated into a risk perceived by the adopter.Behavioural economic biases (e.g., loss aversion) modulate the response of the agents towards a given economic projection of the technology and risk associated with the operation.

En-route ANSPs
ANSPs in Europe adhere to SES regulation and have specific rules for defining their navigational charges following the cost recovery principle.Most en-route ANSPs are publicly funded in Europe, except for the case of the UK, and national security restrictions are still a major issue in controlling the airspace.To provide airlines with some degree of freedom, in our model each origin-destination (OD) pair is covered by several paths, avoiding specific ANSPs in some cases.This adds a layer of competition between ANSPs to the model, which should adapt to provide low charges to the airspace users.Based on the previous approach, the decisions to be computed by the agent are: • Establishment of air navigation charges.It will be highly driven by their cost structure and the traffic projections managed by the entity.Following these projections and the cost recovery principle, they will set the charges to obtain an ideal zero profit, considering the adjustments coming from under and over recoveries from previous years.• Adoption of new technology.New technology is considered to increase ATCO hour productivity.A new technology is thus translated into the need for less ATCOs in operations for the same level of traffic demand.A reduction in personnel does not come at no cost: first, redundancies have a cost associated proportionally to the salary of the employee; second, large dismissals may lead to a strike that has an impact on operational costs.Therefore, a balance has to be found.The set of airports and ANSPs considered form the ATM network of the simulated environment.Figure 2 shows the airports, ANSPs and route network covering each airport pair.

Terminal ANSPs
They are in charge of managing air traffic in the terminal phase of operations (take-off, approach and landing) in specific Terminal Manoeuvring Areas (TMAs) or airports.The decisions to be computed by these agents are similar to the en-route ANSPs: • Establishment of air navigation charges, based on a cost recovery system.The cost recovery system includes a part of cost of capital, by which the contracted companies may obtain benefits.• Adoption of new technology, in order to increase productivity and reduce determined costs.• At a tactical level, airlines react to changes in ANS unit rates by changing the path followed between an airport pair, if that reduces the operational costs.

Industry (technology providers)
The industry (technology providers and aircraft manufacturers) is the main provider of the hardware, software and maintenance of the systems that provide air traffic services in the European ATM.The market is mainly territorially monopolistic (ANSPs), in a way that each ANSP partners with some specific technology provider that creates ad-hoc solutions for their ATC challenges.In spite of the fragmentation in specifications, due to business interests and SESAR efforts new products tend to be interoperable between the European ATM technology adopters.This agent in the model releases the technology to the market when it is developed.Further mechanisms such as pricing strategies and interactions in the development phase have been not considered.

Labour unions
National, European and international organisations support and protect the rights of different guilds critical for the correct functioning of the ATM system, such as Air Traffic Controllers (ATCOs), airport handling staff, and pilots.This agent may decide to call for a strike when an adopter dismisses employees above a given threshold number.The cumulative number of strikes changes the behaviour of the agent involved, making it more reluctant to dismiss staff in the future (learning process).

Regulator
Supra-national, multi-stakeholder European actors (e.g., EUROCONTROL, EASA, SESAR Joint Undertaking, SESAR Deployment Manager) and national governments are considered to be aligned to provide a concise set of rules applicable in Europe as a whole.Thus, this agent applies the policy measures considered in the study scenario.In addition to that, the agent gets track of the monetary quantities spent or obtained from the application of the policies under study.

Outputs and indicators
The successful benchmarking of regulations and policy measures requires a comprehensive assessment of their impact along different dimensions.To that end, the outputs of the model selected for the analysis are to be representative of the performance of the European ATM system in the situations tested.This project aims to be consistent with previous literature, in particular with ICAO Performance Framework [9], the SES Performance Scheme [10] and the SESAR Performance Framework [11].
The following Key Performance Areas (KPAs) and associated indicators have been considered: • Technology adoption effectiveness: Time for adoption and Market share of each technology.
• Economic outcome: Social welfare, understood as the sum of consumer and producer surpluses (en-route and terminal ANSPs, airlines, airports, passengers, and government) plus the externalities (e.g., CO2 emissions), En-route ANSP surplus, Terminal ANSPs surplus, Airline surplus, Airport surplus, Passenger surplus and Government surplus (includes cost of regulation, financial control, subsidies, etc.).• Operational efficiency: Average fuel burnt per flight.En-route throughput per unit time, TMA throughput per unit time, Average union-wide Determined Unit Cost (DUC) for en-route ANS and Average union-wide DUC for terminal ANS.

Calibration and validation
Participatory simulation was chosen as the validation method because it enables the acquisition of data from an environment closely related to a real-world setting.Experts representing the main adopters (i.e., airlines, airports, ANSPs) participated in several sessions, giving answers that enabled three independent validation experts to assess the validation.The results showed that the model behaves in a very realistic manner from the perspective of all three agents, even when faced with extreme values and outliers.For a complete view of the validation experiments, the reader may refer to [12].

Policy assessment scenarios
To gain insights on the conditions for a certain policy measure to be effective and to showcase the capabilities of the tool developed, the agent-based model has been used to test different combinations of ATM technologies and policy measures.First, we investigate why some past technologies achieved high acceptance rates while others did not, and whether certain policy

Past Technologies
The graph in Figure 3 shows the difference in final market share obtained between the scenario analysed and the reference scenario (do-nothing scenario).
It can be seen that this combination of policies produces a faster adoption rate when compared to the reference scenario.In particular, the improvement is larger for the ANSP agents.The reason is that airports already achieved high market shares of the technologies under study, such as A-CDM.On the operational side (Figure 4), the policies effectively achieve reductions in Determined Unitary Costs (DUC) for both terminal and en-route ANSPs with almost no change in the other operational indicators (fuel consumption, en-route and terminal throughput).
It is worth noting that a reduction in DUC is associated with a reduction in unitary ANS costs, leading to lower air navigational charges for airspace users due to the cost recovery scheme followed by ANSPs.
Figure 5 shows the economic indicators.The results show a generalised reduction of both consumer and producers' surpluses.The effect is particularly negative for the profit maximisers (airlines and airports) and the passengers, who suffer higher prices and lower demand satisfied.
The reason for this is that, during the simulation time span, the technological and operational benefits do not completely compensate for the investment costs, leading to a slight decrease (-0.3%) in overall social welfare

Future Technologies
Figure 6 shows a clear positive effect on technology adoption rates across all stakeholders, especially the ANSPs, with respect to not applying the policies proposed.
The market share improvements in this case are higher since the SAFEDRONE concept, contributing to the U-space concept, does not impact on airlines and the full set of stakeholders are not represented.Moreover, the full range of benefits and costs could not be characterised from the available literature.The total costs were excessively large for the benefits, leading to almost no adoption without subsidies.The operational improvements are summarised in Figure 7, where a slight reduction of fuel burnt and considerable improvement of en-route DUC are observed.
Terminal ANSPs are adversely affected in throughput (-0.33%) and DUC (+0.1%).The relative values are small, so more research is needed to explore the operational impacts of the policy measures Social welfare, which aggregates all the surpluses of technology adopters, passengers, and government, experiments a 5% improvement, which is a considerable increase given that the ATM market is small compared to the whole aviation industry.
Figure 8 shows the effect on each stakeholder: the ANSPs are the primary beneficiaries of the measure, airlines improve their profits, and passengers experiment almost no change, while airports and the regulator reduce their surpluses.The regulator surplus is necessarily negative since the subsidies are accounted for as costs for them.

Conclusions
The representation of the main actors in the European ATM industry makes it possible to capture general trends and mechanisms inherent to the system.Validation through a set of participatory simulations, in which the participants and experts that analysed the results concluded that the individual and global behaviour of the agents involved is represented as they would expect, has helped ensure plausible results.In the future, the proposed simulation framework can help optimise the design of policies aimed at accelerating technology adoption, thus contributing to the achievement of SES performance goals and, eventually, helping in the obtention of more sustainable and efficient air traffic.Finally, it shall be noted that the outcome of specific policy combinations is highly sensitive to the benefits and costs of the technologies under study.The cost-benefit analysis (CBA) of new technologies is therefore an essential input to design an optimal policy mix.

Figure 1 .
Figure 1.Model overview Only scheduled airlines have been considered in the model, since they represent most of the IFR flights in Europe.Airlines are private entities, looking for maximising their profit.The main difference in behaviour between them comes from considering their cost structure, which divides them into legacy airlines and low-cost carriers.Airline operations have been modelled so as to follow these steps:• Adoption of new technology.Airlines are the main beneficiaries of ATM improvement, which may reduce fuel costs (e.g., optimising flight paths) or delay times, at the cost of aircraft retrofits and training.As the other technology adopters, airlines have different behavioural biases that modulate their adoption decisions.•At a strategic level, airlines select the number of flights to schedule in each route, depending on costs and traffic projections.The flights offered and the ticket fares are calculated by each airline aiming at maximising their profits, employing a Cournot equilibrium model, which means that the final demand captured from the overall passenger demand depends on the ticket price selected by each airline in the competition.Price-demand elasticities related to airline ticket fares are taken from[8].

Figure 2 .
Figure 2. En-route ANSP's charging zones, airports and routes considered in the model measures could have led to a different outcome.Then, we use the model to forecast the adoption of a set of new SESAR Solutions under a selected policy scenario.The two scenarios considered are: • Past Technologies.The technologies considered for this case study are: Airport Collaborative Decision Making (A-CDM), Continuous Climb and Descent Operations (CCOs and CDOs), Remote Towers and Automatic Dependent Surveillance-Broadcast (ADS-B).Since those technologies were deployed in the past, the simulation time span selected is 2003-2021.The policy mix investigated is Cost-plus pricing + Best equipped best served.• Future Technologies.The technologies considered for this case study are: System-Wide Information Management (SWIM), Dynamic Airspace Configuration (DAC), SAFEDRONE and Initial Trajectory Information Sharing (i4D).Since they are in development and deployment, the simulation time span selected is 2022-2040.The policy mix investigated is Cost-plus pricing + Demonstration projects + Subsidies.

Figure 3 :
Figure 3: Past technologies -Technology adoption market share difference with respect to reference scenario.

Figure 4 :
Figure 4: Past technologies -Operational indicators difference with respect to reference scenario.

Figure 5 :
Figure 5: Past technologies -Economic indicators difference with respect to reference scenario

Figure 6 :
Figure 6: Future technologies -Technology adoption market share difference with respect to reference scenario

Figure 7 :
Figure 7: Future technologies -Operational indicators difference with respect to reference scenario.

Figure 8 :
Figure 8: Future technologies -Economic indicators difference with respect to reference scenario