Future Passenger Air Traffic Modelling: A
theoretical Concept to integrate Quality of Travel,
Cost of Travel and Capacity Constraints
Robin Ghosh, Katrin Kölker, Ivan Terekhov
Research Engineers, German Aerospace Center (DLR), Blohmstraße 18, Hamburg, 21079, Germany,
Phone: +49 531295 3824, Mail: robin.ghosh@dlr.de
5.6.2015
Abstract
In order to model possible future evolutions of the global air transportation system (ATS),
not only the assumptions on external socio-economic conditions are relevant to estimate the
future realized air passenger demand on city pair level. The internal scenario concerning the
air transportation system (ATS), i.e. how the ATS is changing over time with the introduction
of new technologies or new operational concepts has a non-neglectable feedback on realized
demand. Also, the effect of infrastructural changes concerning airport and airspace capacities
have to be considered. That means, the forecast of realized demand and thus realized air
traffic depends equally on the supply offered by the air transportation system. There are
three essential kinds of feedback on realized air passenger demand and air traffic, the supply
provided by the ATS may have. We present a theoretical concept to integrate these three
kinds of feedback (1) quality of travel, (2) cost of travel, and (3) constraints of the supply
side of the air transportation system on modelling global future scheduled air traffic. The
4-layers approach of modelling future evolutions of the ATS is a basic prerequisite to model
these feedbacks, especially the consistent derivation of an aircraft movements network with
information on aircraft generations.
Introduction: Generic build-up of the future ATS
In the DLR project "WeCare" climate mitigating effects of operational and technological changes
are investigated in the context of the future air transportation system (ATS) on a global scale
with a time horizon until 2050. Therefore, at first, a generic model forecasting future air traffic, on
network and fleet basis, is required. This will be implemented in a modular environment, called
AIRCAST (air travel forecast), including 4435 cities worldwide. A generic approach as depicted
in Figure 1 is necessary to assess a multitude of possible changes from the introduction of a single
technology to growth of air travel demand. Pure passenger aircraft fleet models used to assess the
global climate impact of aviation and the introduction of new technologies as in [2], [3], [4], [5], and
[6] have no spatial quality. This is why we combine a fleet scenario analysis with the modelling of
global ATS network evolutions. The spatial distribution of flights is relevant to assess the climate
impact of the ATS and the evaluation of potential mitigation strategies and revolutionary new
concepts. The climate impact of aviation highly depends on the amount, species, altitude and
latitude of emission.[7] We introduce the 4-layer philosophy for a generic build-up of the passenger
air traffic system of the future. A similar approach to decompose the air transportation system
in general and the application by analyzing data of the US air transportation system without a
forecasting methodology can be found in Bonnefoy et al. [8]. Bonnefoy et al. emphasize the
importance of not only analyzing present networks, but also the future structure of ATS networks.
The research conducted in building the AIRCAST environment focuses on the structural evolution
1
Origin-Destination
Demand Network
Routes Network
Aircraft Movements
Network
Trajectories
Network
Infrastructure
Aircraft
Airlines
Stakeholder-Network-Interaction
Disaggregation
Increasing Relevance for Technology Evaluation & Requirements Deduction
Increasing Relevance for Scenario Development
DECOMPOSITION
AGGREGATION
cross-sectional demand
forecast at time slices
GDP
External Scenario Factors
Pop
time series
regional level
country level
city level
Socio-
economic
Scenario
Demand Passenger Kilometers
Revenue Passenger Kilometers
Available Seat Kilometers
/ Directness Factor
/ Seat Load Factor
year
GDP
Population
cross-sectional network
generation @ time slice
Figure 1: Generic build-up of the future ATS in 4 layers [1]
of global air passenger and aircraft networks until 2050. One goal is the derivation of implications
by structural changes of ATS networks on shifts in global mission range frequency distributions
and the proportional shift of deployed aircraft sizes over time. The four layers (see Figure 1)
consist of (1) the origin-destination (OD) demand network, (2) the routes network, (3) the air-
craft movements (ACM) network, and (4) the trajectories network. Each lower layer builds on
the information on the above layers. While the first and the second layer forecast passenger flows
purely from a passenger perspective, the third and forth layer simulate aircraft movements. By
demand here, we mean actual realized demand as described in [9], since our forecasting algorithms
are based on global ADI-data
1
, which may be interpreted as realized demand. Thus, realized
demand are passengers who intended to fly and actually flew in a given year from one origin to a
destination because the travel conditions (e.g. time, airfare, number of transfers, etc.) were right
for them. Our model environment forecasts at first an undirected demand network on city level
(in contrast to airport level), which may be called "city pair air passenger demand". Starting with
the theoretically ideal demand network gradually more information concerning the "operational
reality" of aircraft deployment is included, e.g. effects of hub structures, categories of aircraft
used on segments, and airspace-related trajectory inefficiencies. With the advancement in the
derivation process to each lower layer, there is an increasing relevance for technology evaluation
and derivation of requirements for new concepts. Interfaces with conceptual aircraft design can
be found on the aircraft movements layer concerning the future evolution of flown distances and
the number of flights by aircraft sizes. The trajectories network has an interface with concep-
utal aircraft design concerning flight times, fuel consumption and emissions. On the other hand,
there is an increasing relevance for scenario development when results and assumptions are more
aggregated. This will allow assessing the growth of the ATS against efficiency and technology
improvements dynamically until 2050 on a network and fleet basis. In addition, feedbacks of al-
terations of the supply side of the ATS on the estimation on realized demand are incorporated in
the ATS modelling concept. The overall theoretical framework of systems design, consisting of a
systems analysis and a concept design part, with an inherently quantitative philosophy of scenario
development has been published in [10].
1
Sabre Airport Data Intelligence
2
The combination of dynamic network and fleet evolution information is considered to be highly
valuable for systems analysis, technology assessment, policy planning, and master planning of
aviation infrastructure. The scientific results of the AIRCAST environment and the associated
methodologies in development are especially interesting for [1]:
• market forcasts of manufacturer
• strategic planning of airlines (insight in the dynamics of realized demand on city pair level)
• airport and airspace capacity evaluation
• aviation-related policy making of city governments, national governments, international or-
ganizations (UN, ICAO, IATA)
• dynamic global climate impact assessment of aviation, especially the impact of non-CO2
climate change agents
Exogenous socio-economic scenarios
The AIRCAST 4-layers approach starts with exogenous socio-economic scenarios from external
insitutions as does the AIM
2
modelling approach [11]. The exogenous socio-economic scenarios are
available at different aggregations: global, regional or county level. For the modelling approach
of the AIRCAST environment time series of the parameters GDP and population are required on
city level as an essential input for the city pair air passenger demand forecasting model, named
D-CAST, which directly forecasts demand networks on city pair level in time slices every five years.
Here, a specific breakdown methodology for GDP and population was developed and incorporated
in the tool CITYCAST. The socio-economic inputs used (see Figure 2) are self-consistent scenarios
that rely on global system dynamics models. The data is publicly available. AIRCAST uses the
forecast published by Jorgen Randers "2052" [12] and the five scenarios of the International Futures
Global Modeling System (IFs) [13] as main inputs.
variable
Randers
IFs
GCAM
Randers
Sustainability First
equity, environment,
transparency
Security First
rich, national, regional
Policy First
strong top-down policies
Markets First
maximum economic growth
Base
extension of history
as reference only
Figure 2: Socio-economic scenarios based on system dynamics models serving as input for AIR-
CAST
2
Aviation Integrated Modelling Project
3
The Randers scenario is valuable since the methods and assumptions of forecasting GDP and
population are described with great detail. The results and assumptions are discussed in [12] and
assumptions critically analyzed and extended with useful qualitative scenario information. The
CITYCAST tool further allows a country specific manipulation of the Randers scenario concerning
for example birth and death rates or urbanization factors retrieved from United Nations scenarios.
Additional desirable information would be a an oil price development which is consistent with the
scenarios. This is given for the IF-scenarios, but not for the Randers scenario. The Global Change
Assessment Model (GCAM), developed by the Joint Global Change Research Institute (University
of Maryland) is integrated as a reference and for comparison purposes to other environments that
use this socio-economic scenario for climate change assessments. GCAM is not included in the
AIRCAST environment for forecasting future ATS networks because essential information for the
methodology to decompose those scenarios to city level is not available. A scenario capability
exists for 4435 cities in 215 countries worldwide, meaning that required forecasts on city level of
socio-economic parameters are available in addition to the availability of ADI data in the base
year 2012. 574 cities in 12 countries existing in the 2012 ADI data set could not be included in
the CITYCAST model because of missing socio-economic data. In order to use the time series
of socio-economic parameters more effectively for city pair demand forecast algorithms and to
handle calculation times, a cluster dynamics methodology of cities has been developed. The cities
included in AIRCAST have been grouped in 9 clusters, and thus 45 cluster pairs. The socio-
economic condition of a city in terms of GDP, population, and GDP per capita defines its cluster
membership. The cluster membership is calculated for each time slice every five years according to
its changing socio-economic conditions over time. [14] The cluster dynamics of cities is important
to increase the precision of forecasting a global city pair demand network.
Demand Network
The forecast of the air passenger demand network in future time slices follows two steps as depicted
in (Figure 3): (1) topology forecast and (2) forecast of the number of passengers on an edge.
routes network
aircraft movements network
trajectories network
first predicted time slice
time slice 2050
TOPOLOGY-FORECAST
Evolution of the demand network topology over
time every five years
PASSENGER-FORECAST
Evolution of the number of passengers on city
pairs over time every five years
origin-destination demand network
quality of travel
frequency
travel time
number of
transfers
cost
new
technologies
direct and
indirect
operating cost
exogenous socio-economic scenarios
first predicted time slice
time slice 2050
Figure 3: A two steps process of forecasting air passenger demand evolution: topology and number
of passengers on city pairs worldwide
4
Since we forecast an undirected network which does not know the direction of origin and des-
tination, we call one connection demand city pair. A detailed approach to the applied demand
forecasting methodology can be found in [9] and [15]. The basic demand network topology is de-
fined by the data of the base year 2012. The step of topology forecasting estimates the appearance
of new demand connections for each time slice using weighted similarity-based algorithms accord-
ing to a socio-economic scenario. Subsequently, passengers on the defined demand connections
(edges) are forecasted. We defined metrics to measure the quality of travel in order to quantify
feedbacks of the introduction of concepts like flying slower or Intermediate Stop Operations (ISO)
on realized demand. Feedbacks on realized demand due to alterations by the introduction of new
aircraft with a given aircraft price and a given fuel efficiency or airline business model trends
resulting in changing direct and indirect operating cost need to be modeled through a detailed
estimation of airfare.
Routes Network
The routes network is forecasted based on the previously generated air passenger demand network
in time slices on city pair level. In order to model the global routes network every city pair of
the demand network and the associated weight on an edge in terms of passengers is analyzed
individually. That means, the routes passengers take, are modelled for each demand city pair
seperatly. The probabilities for routes are defined by analyzing historical ADI data. Figure 4
depicts the overall process.
City A
City B
city pair demand
(1) Definition of possible /reasonable routes for each
demand city pair worldwide
(2) Modelling probabilities of passengers choice
for each possible route
City A
City B
p = 0.15
City A
City B
100.000 PAX
50.000 PAX
35.000 PAX
15.000 PAX
(3) Allocating amount of passengers according to
route probabilities for every demand city pair
worldwide (more than 500.000 per time slice)
p = 0.5
p = 0.35
(4) Aggregation from city pair routes to passengers
on segments worldwide
Figure 4: Process of forecasting a routes network: deducing passengers on segments
5
![Figure 5: Forecasting aircraft movements on segments [16]](/figures/figure-5-forecasting-aircraft-movements-on-segments-16-39vtqghf.webp)
![Figure 1: Generic build-up of the future ATS in 4 layers [1]](/figures/figure-1-generic-build-up-of-the-future-ats-in-4-layers-1-2zo23ajz.webp)
![Figure 6: Forecasting an aircraft movements network with aircraft generation information: Interfaces between the aircraft movements layer and the fleet renewal model FFWD [3]](/figures/figure-6-forecasting-an-aircraft-movements-network-with-1r6fb8me.webp)
