Fourteenth USA/Europe Air Traffic Management Research and Development Seminar (ATM2021)

Probabilistic Complexity in support of Airspace

Capacity Management Optimisation
Integration of Complexity Management and Dynamic Airspace Configuration

Danlin Zheng Zhang, Eva Puntero Parla, Marta Sánchez Cidoncha

CRIDA, A.I.E.
Madrid, Spain
mscidoncha@e-crida.enaire.es

Abstract— The present paper addresses innovative ATM solutions configurations to accommodate as much as possible the demand
for Demand and Capacity Balance, in particular Dynamic and the airspace users’ optimal trajectories.
Airspace Configuration. Capacity Management processes in these
environments require more critically the use of support
complexity metrics tailored to the relevant use cases. An Enhanced
Complexity Management including both the improvement of the
metrics introducing demand uncertainty and their integration into
the overall process is presented. The assessment of this Enhanced
Complexity Management is presenting, assessing its technical and
operational feasibility and showing its improvements in capacity
and cost-efficiency.
Figure 1. Traditional vs. DAC Airspace Structure (proposed for
Keywords - Cognitive Complexity, Uncertainty, Capacity
Spanish Sectors in SESAR 2020 PJ08-W1 AAM Real Time Simulation
Management, Dynamic Airspace Configuration exercise )

I. INTRODUCTION In order to enable DAC feasibility [3], the determination of

In Air Traffic Management the concept of Dynamic the capacity for all these sectors built from elementary volumes
Airspace Configuration (DAC) pursues higher levels of is yet an open question.. This is due to the fact that capacity of
flexibility in the airspace design and configuration process. It traditional sectors is established in terms of ‘entry counts’
aims at increasing the capability of the airspace to adapt to the (number of aircraft entries in the sector per time unit), whose
traffic demand in nominal situations as well as in case of maximum value is dependent on the operational knowledge
unexpected events and demand. The concept [1] proposes to about the sector of the relevant Air Traffic Controllers (ATCo)
leave traditional pre-defined airspace structures behind and opts and other operational staff. However, the use of non-predefined
for a dynamic configuration of the air traffic control sectors, airspace structures requires new methods to determine sectors’
which can be continuously adapted to the traffic evolution. DAC capacity with a higher support of automation and accounting for
sector design and configuration process is based on the use of the quantification of complexity factors affecting the provision
elementary volumes of airspace, that are the building blocks to of the ATC service in a DAC operational sector. One risk of this
consolidate the sector configuration plan most suitable to the approach is a high risk of capacity planning instability if demand
expected demand. This process runs from the strategic planning uncertainty is not properly taken into account during the sector
(weeks before the day of operation) up to the execution phase, design and configuration process.
that is, up to 20 minutes before the final sector configuration is Probabilistic Cognitive Complexity (PCC) assessment
implemented. stands as a solution for addressing these two issues. Air traffic
Figure 1 compares the results of a traditional airspace design complexity is a measure of the implicit difficulty for an ATCo
vs. a DAC one [2], where additional airspace volumes are used when managing traffic in his/her area of responsibility (airspace
to allow higher flexibility in the consolidation of operational sector). The consensus view among the ATC research and
sectors. operational communities is that complexity drives controller
workload, which in turn is thought to ultimately limit sector
From the simulations and tests performed, it is foreseen that capacity. PCC is a complexity metric which biggest asset is that
the deployment of DAC airspace design process will multiply it considers complexity factors intrinsic to traffic and sector
the possible options of airspace configurations of operational shape to derive their effect on Air Traffic Controllers’ (ATCo)
sectors and will redound in higher capacity and more flexible cognition, all of this without underestimating the uncertainty of
the traffic demand on complexity prediction.
 PCC is a key enabler to assure the effectiveness of DAC by 2) Interaction between a climb/descend flow with a cruise
providing a more accurate estimation of capacity that goes flow;
beyond ‘entry counts’. This capacity estimation will allow a 3) Interaction between two climb or descend flows;
better prediction of actual overloads improving ATCo 4) Interaction between a climb flow with a descend flow.
productivity and ensuring a cost effective planning of the
airspace configuration. Moreover, the higher reliability of the
capacity estimation will allow reducing capacity buffers needed
to ensure safety, thus increasing the capacity of the airspace.
This paper presents the enhancements brought by PCC
assessment to DAC short-term1 Capacity Management
(CM), based on an assessment through Fast Time Simulation
(FTS) techniques focused on the evaluation feasibility, capacity
and cost-efficiency of the process and resulting airspace and
traffic picture. The integration of PCC into DAC CM includes Figure 2. Interaction between Two Traffic Flows
the development of the mathematical formulation of Cognitive
Complexity (CC) [4] to best fit DAC CM requirements as well Potential conflicts (X2)
as the review of the DAC CM use cases to ensure process
effectiveness. Specifically, this paper presents the following: For each potential conflict (“trajectories” crossing) between
a pair of flights at time t, the following formulae is used to
 Description of Cognitive Complexity mathematical calculate the complexity due to the conflicts:
formulation and the introduction of traffic demand time
uncertainty to obtain PCC evaluation (section II).
𝑋 (𝑡) = (𝑌 · 𝑌 · 𝑌 · 𝑌 · 𝑌 )  
 Enhancement of DAC use cases to integrate the use of
PCC assessment, specifying decision-making Being n the number of conflicts at time t, Y1 the difference
processes, required automation support and human in flight level of both flights, Y2 the difference in time over the
interaction (section III). crossing point, Y3 the distance to the potential conflict, Y4 the
crossing coincident with a pre-identified crossing-point and Y5
 Summary of experimental results obtained through the relative vertical speed of both flights.
simulation and discussion feasibility, capacity benefits
and cost-efficiency of the solution (section IV). Number of flights in evolution (X3)
Number of flights in evolution within the sector at time t.
II. PROBABILISTIC COGNITIVE COMPLEXITY
Number of flights out of standard flows (X4)
A. Cognitive Complexity due to Traffic Demand Number of non-standard flights within the sector at time t.
Cognitive Complexity (CC) [4] metric estimates controllers’
mental workload through a function combining important In order to give the metric an easily interpretable value,
abstractions (parameters) related to the complexity of the traffic: Cognitive Complexity is normalized to allow its comparison to
an ISA scale. The ISA (Instantaneous Self-Assessment) method
Flows interactions (X1) uses a special 5-key level at each working position [5] being 1
the smallest and 5 the greatest complexity. Several functions and
For a pair of flows at time t, the interaction between them is
techniques have been tested to link up the aforementioned CC
computed by the following formulae:
parameters with corresponding ISA values. The best correlation
𝑋 (𝑡) = 𝑜𝑐𝑐1 · 𝑜𝑐𝑐2 · 𝑊  between them was obtained by a mix function given by the
following formulae:
where Occn is the number of flights from Flown that are
contained in the interaction space between both flows at time t.
W is the weight considered for each type of interaction (see 𝑓 = 𝐶𝐶 (𝑡) = 𝑎 +𝑏 ∗𝑋 +𝑐 ∗𝑋 +𝑑

Figure 2). A flow is defined as a group of aircraft that has similar
flight path and aircraft performance within a specific airspace. ∗ 𝑒𝑥𝑝(𝑒 ∗ 𝑋 )
The clustering of the aircraft can group the flights into different Where N is the number of parameters, ai, bi, ci, di and ei the
flows. The standard flows are the ones that represent the constants and coefficients and Xi the parameter i.
majority (over 70%) of the studied traffic sample. The rest of
flights not assigned to any of the main flows are considered as B. Cognitive Complexity due to Sector Shape
non-standard flights. There are four types of flows interactions The Cognitive Complexity algorithm involving Sector
(ordered from less complex to most complex): Shape (SS) factors reflects the spatial complexity of the
1) Interaction between two cruise flows; sectorization in relation to the traffic flows. These factors are
identified as key variables to be taken into account when DAC

1
DAC short-term Capacity Management processes are those taking place on
the day of operations up to 20 minutes before execution time.

This paper presents research conducted within COTTON project (Grant

783222) funded by the SESAR Joint Undertaking as part of SESAR 2020
Exploratory Research Programme within the framework of the EU’s Horizon
2020 research and innovation programme.
solution is implemented. These factors are considered within the
complexity assessment as: 1
𝐷 = (· 2 𝑖𝑓 𝑡ℎ𝑒 𝑆 𝑖𝑠 𝑙𝑒𝑠𝑠 𝑡ℎ𝑎𝑛 10𝑁𝑀) 
𝑠
Angles and Vertices (D1):
This factor accounts for the convexity to minimize the Being j the number of crossing points and Si the distance of
number of re-entering flights. D1 is the sum of number of the each crossing point to the closest sector border. If the distance
vertices and the number of vertices whose angle is less than 90°. between critical point and sector boundary is less than 10NM,
The more vertices of this type are present in the sector, the higher there will be less reaction time to resolve possible conflict that
is the complexity since it is more likely that re-entries occur (see occurs in such critical point, thus the parameter’s relevance is
Figure 3). doubled.
The algorithm concerning sector design factors for DAC
process is given as follows:

𝑓 = 𝑆𝑆 = 𝑒 ∙𝐷  

where ei is a weighting coefficient to homogenize the

dimensions and weights of the four Di factors accounting for
Figure 3. Example of D1 Factor sector shape complexity to be comparable with the CC values.
This SS value is a static value per sector.
Flows Orientation and Position (D2):
C. Uncertainty Quantification for Probabilistic CC Modelling
This factor accounts for the orientation and position of main
flows present in the sector to minimize the number of The uncertainty is one of the most influencing factors on the
coordination (see Figure 4). D2 takes into account as main demand forecast, thus, on the demand and capacity balancing
complexity parameters the centralism and the parallelism of the (DCB) process. The input to all the four parameters in Cognitive
flows with respect the sector borders. The formulae for this Complexity are the trajectories of the flights that are going to
factor is: cross the sector of interest along the time period for complexity
calculation. The trajectories used for the prediction of
𝐴1 − 𝐴2 complexity at a given time horizon (before the execution time)
𝐷 = + |𝜇 − 𝜇 | + |𝜇 − 𝜇 |   are based on flight plans (FP): either available (already filed by
𝐴1 + 𝐴2 Airspace Users) in the FP management system at the target time
horizon or created from historical data. The trajectories are built
Being j the number of main flows within the sector, A the
from these FPs assuming the shortest path between two way-
area at the left/right side of the average flow track and µ the
points.
angle at the left/right side of the average flow track with respect
to the sector boundary (four angles for each flow). These predicted trajectories are uncertain, and the actual
execution of flights will differ from the plans:
1) In times over the expected points;
2) In the 2D route flown (flown 2D trajectory way might be
different from the planned one);
3) In the flight’s evolution (planned flight levels achieved
later or sooner than planned, differences in climb or descent
rates, etc.)
Figure 4. Example of D2 Factor Other uncertainty sources may be important such as ATCO’s
instructions for traffic separation or even the weather forecasts.
Sector Vertical Size (D3): Given that the focus is on the DAC Short-Term tactical phase,
one of the most influencing sources is time variation. A first
This factor accounts for vertical size of the sector to ensure
intent is to take into account the time uncertainty among other
a minimum flight crossing duration and a minimum volume to
uncertainty sources into the complexity metric and improve this
resolve the potential conflicts. D3 is the inverse function of the
methodology by incorporating more uncertainty sources in
number of Flight Levels (FL) present in the sector. The bigger is
further researches. This means that uncertainty will be
the number of FLs the less complex is the sector.
incorporated in the planned trajectories as variability on the
1 Estimated Time Over (ETO) the planned points within the
𝐷 =   sector of interest. Since the uncertainty related to time error is
(𝑁º 𝑜𝑓 𝐹𝐿𝑠)
only applicable to traffic parameters, then the uncertainty is only
Crossing Points Distance (D4): incorporated into the CC metric, not into the SS metric (fixed
This factor accounts for the distance between the sector value for each sector).
boundary and the pre-identified crossing points or major areas
of interaction. The mathematical formulation is given as:
 To obtain the variability in ETO for each flight at the sector sector but that considering their ETOs and their corresponding
of interest, it is assumed the following: temporal error assigned depending on the belonging flow they
 Assumption 1: The variability of ATO (Actual Time can be inside the sector at time t.
Over) with respect to ETO for a flight at any point within 2) Once the set of flights to be considered in the calculation
the sector is equal at the Entry point of the flight to the is identified, the calculation of the CC is performed N times:
sector.  N is the number of runs (or required sample size) of the
Monte Carlo simulation needed to achieve results with
 Assumption 2: The variability of ATO with respect to statistical validity;
ETO at the Entry point of a flight to the sector is the
same for all flights within the same flow inside the  In each run, the temporal error of ETO for each flight of
sector. the set is chosen randomly according to the form of the
corresponding probability distribution of error;
Following these assumptions, the aim is to characterize the
variability of ATO with respect to ETO at each entry point for  The flights are re-positioned according to the error,
each flow, that is, the sector (in this case the DAC area) entry meaning that some flights might be positioned out of the
time error. For flights out of standard flows, the characterization sector and some that were outside nominally to be inside
is made for the whole set of non-standard flights as a unique non- the sector;
standard flow. The characterization uses a representative traffic
sample of the sector of interest (one month) and is done for the  The four parameters of the Cognitive Complexity (see
FPs available at different time horizons: 12 hours, 3 hours, 1 section II.A) are calculated taking as input the flights
hour and half an hour before actual entry time of the flight (see within the sector according to the re-positions;
T-12, T-3, T-1 and T-0.5 in Figure 5).  Each run results in a value of Cognitive Complexity for
For each flow, it is built a chart counting the number of times time t.
that a certain error (ATO – ETO) occurs. The Figure 5 represents 3) After N runs, there are N values of Cognitive Complexity
the time uncertainty evolution (X-axis time error in minutes and for time t. From this set of values, it is built a probability
Y-axis frequency of each time interval) along forecast horizon
distribution of the Cognitive Complexity where each value has
(being T the time of actual entrance) within the DAC area for a
specific main flow. It can be seen that as time approaches to the its associated probability of occurrence and from which the
time of execution, the shape of the distribution becomes more cumulative probability of Cognitive Complexity being below or
regular/symmetric and centralized to zero minutes, which is as above certain value can be calculated (percentile).
expected since uncertainty is reduced.

Figure 6. Probability Distribution of Cognitive Complexity at time t

from Monte Carlo Simulations

III. AIRSPACE CAPACITY MANAGEMENT OPTIMISATION

Figure 5. Occurrence of (ATO-ETO) at different time horizons for a
specific main flow within DAC area A. Enhanced CM Process
DAC short-term Capacity Management processes are
In order to integrate this uncertainty model into the CC proposed to be improved by integrating the use of the presented
assessment, Monte Carlo simulations are used to provide PCC PCC, when the demand accuracy allows a reliable estimation of
values from the abovementioned entry time error distributions. cognitive complexity. Specifically, supporting an FMP (Flow
Management Position) when performing DCB tasks during the
The output expected is a probability distribution of the CC
day of operations and up to 20 minutes before execution. As part
metric at each instant of calculation. The methodology used is
of these tasks, the FMP will monitor the status of the DAC sector
as follows (see Figure 6):
configuration, identify potential hotspots and propose
1) To build the probability distributions of the CC at t, the alternative sector configuration to solve them.
calculation consider the flights that are nominally within the In the proposed enhanced CM process, PCC assessment
sector at time t plus the flights that are nominally out of the could support FMP in the identification and assessment of
hotspots as well as in the search of appropriate configuration for  Minimisation of number of controllers (open sectors).
each traffic situation. PCC metric will serve to evaluate the
congestion and workability of the sector configurations and  Maximisation of the weighted load balancing over the
identify if the configurations are able to resolve hotspots. The sectors of the configuration.
following paragraphs describe how PCC will support FMP  Minimisation of the average complexity of the sectors
tasks. of a proposed configuration.
First of all, when monitoring sectors’ capacity, the use of  Minimisation of the number of sector difference
PCC will allow the FMP to easily take into account the between consecutive configurations.
workability of the sectors configuration according to their
complexity and the reliability of this information according to As a result, the FMP, supported by the optimisation tool will
the uncertainty of the demand. For example, in the figure below, decide which sector configuration plan better matches the
the hotspot identification is different depending on the PCC demand prediction at this planning phase.
percentile use, so there is a 10% probability of having a 14
minutes hotspot and a 30% probability of a 6 minutes hotspot. B. Probabilistic Cognitive Complexity for Airspace
Configuration Optimisation
The Probabilistic Cognitive Complexity (PCC) assessment
is based on the combined use of the metrics CC and SS,
incorporating the time uncertainty into the CC metric (see
section II). This is done through the integration of the Cognitive
Complexity function (f1) with the Sector Shape function (f2)
resulting into a ffinal cost function, which is called OPT value for
each of the candidate configurations of sectors. The value in
OPT is used to rank the configurations for a predicted traffic
demand at a forecast horizon during the period of analysis. The
different candidate configurations are therefore ordered based
Figure 7. Hotspot identification supported by PCC assessment on the OPT value.
The FMP may decide, according to the prediction of the For evaluating a configuration (structuring the airspace in a
hotspot, its probability and the planning time, that there is a need particular split into sectors), each of the parameters included in
to apply a DCB measure. In case that the FMP wants to evaluate Cognitive Complexity (section II.A) and in Sector Shape
the application of an alternative sector configuration, PCC can (section II.B) are taken into account to build the following
support him/her either by proposing a set of alternative sector metrics at a given time t:
configurations solving the hotspot or providing the assessment M0: Number of sectors for which the probability of CC
for alternative configurations (what-if functionality). To do so, a being greater than the peak overload threshold is above a
sector configuration optimization algorithm, based on PCC given percentage at time t;
metric assessment, aims at balancing demand and capacity
dynamically inside the DAC area. The algorithm uses PCC to M1: Number of sectors for which the probability of CC
assess sector load of each sector configuration and to identify being greater than the sustained overload threshold is above
overload according to the following configurable parameters: a given percentage at time t;
 Peak Complexity Threshold: the maximum M2: Difference between maximum and minimum values of
complexity that is manageable by an ATCo the most probable CC of the set of sectors at time t;
instantaneously.
M3: Average value of the most probable CC of the set of
 Sustained Overload Period: The maximum period of sectors at time t;
time over which it is acceptable to work over a M4.1: Average for the set of sectors of the standard
Sustained Complexity Threshold. deviations of the probability distributions of CC at time t;
 Sustained Complexity Threshold: the complexity M4.2: Mean value of all runs of the mean CC of the set of
value over which it is not acceptable to work a time sectors at time t;
interval longer than the Sustained Overload Period.
M5: Average value of the SS metric for the sectors of a
Once these configurable parameters are set, the algorithm is proposed configuration;
able to rank sector configurations according to the assessment of
a selection of the following criteria (in detail in Section 3.2): M6: Difference between the maximum and minimum values
of the SS metric among all the sectors of a proposed
 Minimisation of overload by avoiding those sector configuration.
configurations presenting one complexity value over the
Peak Complexity Threshold or one set of complexity These parameters are latter used to calculate the different
values longer that the Sustained Overload Period over variables that compose the integrated OPT cost function as
the Sustained Complexity Threshold. follows:
 𝑂𝑃𝑇=ffinal=OPT (f1,f2) = 𝑎·𝑋1+𝑏·𝑋2+𝑐·𝑋3+d·X41+e·X42+f·X5+g·X6 (8) exceeded or whether the sustain threshold might be exceeded for
sustained periods. To do so, the following conditions were
Being, assessed:
a, b, c, d, e, f and g the weight of each parameter X.  If the percentage of time that M0 is equal or higher than
X1 is the difference between average M2 during the period one is more than 5%, then the configuration is
and its minimum value among all the proposed discarded;
configurations.
 If the percentage of time that M1 is equal or higher than
X2 is the difference between the standard deviation of M2 one is more than 80%, then the configuration is
during the period of analysis and its minimum value among discarded.
all the proposed configurations.
In the second step, OPT value is calculated for the remaining
X3 is the difference between the average M3 during the configurations. The configurations whose ‘OPT’ value is lower
period of analysis and its minimum value among all the than the average ‘OPT’ are considered as good candidates for
proposed configurations. implementation since the excessive overload is guaranteed
through the application of the condition assessed in step 1 and
X41 is the difference between the average M41 during the the evaluation of the Probabilistic Cognitive Complexity is over
period of analysis and its minimum value among all the the average value for the remaining sector configurations.
proposed configurations.
In the third step, priority is given to those sector
X42 is the difference between the average M42 during the configurations that, fulfilling the previous requirements, present
period of analysis and its minimum value among all the the lowest possible number of sectors. Therefore, the sector
proposed configurations. configurations that survived to step two are ranked based on the
X5 is the difference between M5 and its minimum value number of sectors. Sector configurations with the same number
among all the proposed configurations. of sectors are ranked based on the OPT value from least to
greatest.
X6 is the difference between M6 and its minimum value
among all the proposed configurations. The optimiser will then propose to the user the best choices
found according to the assessment. In the case of the exercises
The weight of the variables (a, b, c, d, e, f and g) shall be presented in this paper, the tool showed juts the three best-valued
fine-tuned to reflect the business strategy of the Air Navigation sector configurations.
Service Provider (ANSP) or specific airspace region where the
algorithm is used. For example, avoidance of overload can be
prioritised against balancing workload amongst sectors. This IV. EXPERIMENTAL RESULTS
will determine if more weight shall be put on the parameters
measuring the balance of complexity between sectors, sector A. Test Scenarios
shape complexity level or the number or level of cognitive For the evaluation of the Optimiser tool, different scenarios
complexity itself. The calibration of the weights shall be done in to test its validity have been defined. A date is selected for the
two steps: validation (05/07/2017) for a specific time period (19:00 to
20:00), which contains real regulation periods with high traffic
1) Normalize the variables so that they contribute equally demand over a set of selected Spanish sectors (TER, ZGZ and
to the OPT function: CJI). A list of possible sector configurations (up to 30) has been
a) Adjust the weight based on the average value of all identified as input for the Optimiser to choose the most
variables of the cost function of the candidate sector appropriate among them, which means the most suitable for the
configurations. expected traffic demand.
b) A normalized weight will result for each variable in There is a Reference (REF) scenario, which accounts for the
order to make them have the same importance as a basis. baseline of the validation and represents the real situation that
2) Re-allocate the weights for all variables based on happened the day of operation. And there is a Solution (SOL)
scenario, which uses the proposed solution for DAC concept
operational expert judgement. In the exercise presented in
improvement. The goal of this test validation is to obtain initial
this paper, the weights were set to account for an ANSP results on the operational performance benefits that the tool can
performance strategy where the priorities are to minimise the bring to the DCB process. The workflow for the different
overall average complexity value over the balance between the scenarios is described below.
sectors and the CC metric over SS (assuming an environment
Reference Scenario (REF). Cognitive complexity metric is
where traffic contributes more to mental workload than the
chosen for sectorisation selection in DAC reference scenario.
shape of the sector). The input to the tool is composed of the flight plans (DAC traffic
The optimisation process follows then a three step approach: demand) and all the possible sector configurations to be
assessed. The tool computes the cognitive metric of each sector
In the first step, some sector configurations are disregarded under each configuration. Finally, the sector configuration
to ensure that the candidate configurations do not present severe optimiser contains criteria to select the best-fitted configuration
overloads, that is evaluate whether the peak threshold might be as described previously (see Figure 8).
 configurations, together with the human final selection, these
last sector configurations are depicted in Figure 10.

TABLE I. PROPOSED SECTOR CONFIGURATIONS ACCORDING TO

PROTOTYPE TOOL AND HUMAN SELECTION

Forecast Human Number of

SCN Tool Proposal for TOP3
Horizon Selection Sectors
REF 4 8 29 4 3
T0
SOL 4 28 18 4 3
REF 14 29 6 14 3
T-0.5H
SOL 4 11 6 4 3
REF 4 29 11 4 3
T-1H
SOL 8 4 11 4 3
REF 4 8 14 4 3
T-3H
SOL 8 4 11 4 3
REF 29 2 6 29 4
T-12H
Figure 8. Reference Scenario workflow (REF) SOL 8 4 11 4 3

Solution Scenario (SOL). The tool is completed with the It shall be considered that DAC concept allows changing
incorporation of time uncertainty distributions to the input configurations until 20 minutes before the start of the period of
demand based on historical data. Monte Carlo simulations are operation. REF proposes a sector configuration change at T-
executed to obtain the probabilistic distributions of CC (PCC) 0.5H, which is later refused at T0 when CONF.14 is not
for each time t of the period of analysis. The Configuration considered as one of the most optimal. SOL scenario is more
optimiser is adapted to the cost function described previously for stable since the final human selection (based on expert judgment
this scenario (see Figure 9). and experience) is always the CONF.4, like for SOL scenario.

Figure 9. Solution Scenario workflow (SOL)

The tools needed are completed with a fast time simulation

tool (RAMS) to validate the efficiency of the proposed
sectorization through indicators.
Figure 10. TOP 1 sector configurations proposed by the tool2
Note that in the next section of results, for the Capacity
improvement assessment, there will be two different Solution It can be highlighted that CONF.4 is, amongst all considered
Scenarios (SOL1 and SOL2), when assessing the prediction of options, the one that is more similar to the sector configuration
overloads at sector level. This is due to the use of uncertainty that was applied for the period of analysis in the day of
distribution and will be explained in the corresponding section. operations. This proves that the proposed Solution is able to
Whilst for the Cost-efficiency assessment and presentation of propose a feasible configuration from a single set of input traffic
the sector configuration results, it will be at configuration level, equaling the selection based on extensive experience and
then the results are only account for a unique solution scenario, knowledge of the target airspace.
applicable for all probabilistic distributions.
Airspace Capacity Improvements
B. Test Results The number of remaining hotspots has been evaluated in the
The results using the prototype tool evidenced the ability of different scenarios. This indicator measures the number of
the proposed enhanced methodology of finding an optimal overloads (at configuration level) that eventually could not be
sector configuration. Table I shows the result of the optimization resolved with the optimal configuration selected.
algorithm per forecast horizon and scenario. The table includes A remaining hotspot is identified if a peak overload (CC
the results of the cost function as the “Tool proposal for TOP 3” value above 4.7) or a sustained overload (CC value above 3.5

2
The sectors shown are in 2D but the vertical profile is presented in the text
next to each one.
for more than 5 minutes and maximum 1 blank minute) is Table II presents both the prediction of the sustained
computed with flown (simulated with RAMS) trajectory. For overloads (in terms of H/P/F/M rate) at sector level and the
SOL scenarios, when treating with probabilistic complexity, two remaining hotspots (in terms of nº of remaining hotspots and TS
percentiles are chosen to calculate PCC value: 90% as scenario and CS severity values) at configuration level for the final
SOL1 and 70% as scenario SOL2. Percentile 90/70 (PCC1/ PCC2) selected configuration for each scenario at T-12H and T-0.5H
are, respectively, the values of PCC in the resulting probabilistic forecast horizons (the benefits are marked in green with respect
complexity distribution such that the probability of PCC being to the REF scenario). The T-3H and T-1H forecast horizons did
less than or equal to them is 90%/70%. not present any overload at sector level, thus neither any
remaining hotspot at configuration level. Therefore, no analysis
The severity of the remaining hotspots is also considered, can be done from these results.
both in terms of duration and complexity level. Time Severity
(TS) is obtained as the overload period divided by the simulated TABLE II. CAPACITY ASSESSMENT
period (60 minutes). Complexity Level Severity (CS) considers:
level 1 between 3.5 and 4; level 2 between 4 and 4.5; level 3 SCN CONF. Nº Sect.
Suitability Predictability
above 4.5. If more than a sector has a remaining hotspot, the Hotspots TS CS H P F M
average value is considered at configuration level. T-0.5H
REF 14 3 2 0.3 2 0 0 0 2
It is considered an improvement in capacity if a change in SOL1
SOL2
4 3 1 0.2 2
1
0
0
0
0
0
0
1
the selected optimal configuration thanks to the proposed T-12H
Solution leads to better accommodation of the traffic demand, REF 29 4 1 0.1 2 0 0 0 1
SOL1 0 0 0 1
and thus, the remaining hotspots are reduced with regards to the SOL2
4 3 1 0.1 1
0 0 0 1
Reference Scenario. In this case, it is considered that there is an
improvement in suitability of the selected configuration to the At half an hour forecast horizon, the selected configuration
traffic demand. in both solution scenarios is the same CONF.4, whereas in REF
scenario the configuration is CONF.14. With respect the REF
The exercise evaluates as well the capability to effectively scenario, it can be seen that the suitability is significantly
predict hotspots by counting: improved, since the number and severity of the remaining
 Hit (H): a predicted peak or sustained overload, which hotspots are reduced in solution scenarios. With SOL1
matches an actual one, meaning a minimum 50% match (PCC1=Percentile 90%) the overload is hit totally and no misses;
between the time periods of both in the case of sustained. and with SOL2 (PCC2=percentile 70%) the overload is missed.

 Partial hit (P): a predicted sustained overload whose The following figures show examples of visualization of
duration is partially matched (<50%) by the time period of overload prediction at T-0.5H. Figure 11 and Figure 12 depict
an actual sustained overload. CJI sector where there is a sustained overload according to the
actual trajectories. The predicted complexity distribution curve
 False alarm (F): a predicted peak or sustained overload that with percentile 90 hits the same overload period while the curve
does not occur when using actual demand. of the percentile 70% does not forecast any overload.
 Missed (M): actual peak or sustained overload, which is 12 hours before the start time of analysis period, the REF
not predicted when using the input demand. scenario gives the CONF. 29 as the most suitable while both
solution scenarios give the CONF. 4 as most appropriate.
The calculation of the level of hit is exampled as: An actual
sustained overload occurs between 19:19 and 19:31. The In REF scenario, there is one actual sustained overload in
predicted sustained overload estimated is from 19:16 to 19:23. CONF. 29, which was missed in the prediction. SOL scenarios,
The overlapped time period is 5 minutes: from 19:19 to 19:23. with CONF.4, provides a capacity improvement in terms of
The percentage of matching is the overlapping time period remaining hotspots (reduced CS level). However, there is no
between predicted and actual overload divided by the real improvement observed in overloads predictability, as
duration of the overload (13 minutes): 5/13=38%. Thus, this complexity algorithm was not able to predict them (one missed
overlap is identified as a Partial hit (P). overload).
These indicators provide, at sector level, an estimation of the
robustness of the chosen complexity metric (predictability)
facing uncertainty in demand. An improvement is achieved
when the Solution Scenarios’ sectors present better hit, false
alarms and misses rate than the Reference Scenario’s sectors.
For the considered time horizons and the selected
configurations there are no peak overloads in any scenario.
Therefore, the robustness of prediction of peak overload cannot
be assessed. However, the effectiveness of the OPT cost
function is highlighted since this indicator has been considered
as part of the algorithm.
Figure 11. Overload prediction for SOL1 scenario for sector CJI at T-
0.5H
 comparing both options, there is no reduction in staffing costs
but there is a reduction in occupancies per minute and a similarly
(qualitative) balanced workload (see Figure 13 and Figure 14).
The reduction in OCC/min in CONF. 4 is possibly due to a better
match of the airspace configuration to the demand so that more
flights are only flying through one sector instead of two or three.

Figure 12. Overload prediction for SOL2 scenario for sector CJI at T-
0.5H

As a summary, forecast horizons T-12H and T-0.5H present

benefits in predictability and suitability in Solution scenarios
with respect to the Reference scenario. The incorporation of
uncertainty to the complexity evaluation adds stability to the
overload prediction making it very dependent on the percentile Figure 13. CC distribution for CONF.14 at T-0.5H
used to declare a hotspot. i.e. 90% percentile predict better the
overloads than using 70%. However, in forecast horizon T-3H
and T-1H the overload prediction and suitability cannot be
evaluated in the analysed scenarios as there were no remaining
hotspots.
Cost-Efficiency in Airspace Configuration
A benefit is considered achieved in Cost-efficiency if the
sector configuration proposed by the proposed Solution:
 Implies no reduction in the number of sectors, but there
is a reduction of the average occupancies per min
(OCC/min) and/or a more balanced CC among sectors.
 Implies a reduction in the number of sectors and still the
average OCC/min and balance of CC are at a
Figure 14. CC distribution for CONF.4 at T-0.5H
manageable level.
In Table III, the results of this analysis are shown per time At T-12H, SOL scenario proposes a more cost-efficient
horizon and scenario. The benefits are marked in green and the sector configuration since it implies a reduction in the number
worse in orange with respect to the REF scenario. In the T-3H of sectors and still the average OCC/min is below a manageable
and T-1H forecast horizons the proposed sector configurations level (with an increase of 0.5 flights in average with respect to
(human selection) remain the same for both REF and SOL the REF scenario). The workload balance is still under a
scenarios, thus no reduction in number of sectors nor average manageable level, even reducing the number of sectors (see
OCC/min at configuration level is observed. Therefore, no Figure 15 and Figure 16).
analysis can be done from these results. Note that for Cost-
Efficiency assessment, analysis is at configuration level, thus the
probability of CC is not used in this case. Deterministic CC
values are shown in the following graphs to compare more
simply the CC balance among sectors with respect the REF
scenario.

TABLE III. CAPACITY ASSESSMENT

SCN CONF. Nº of Sectors Nº of ATCo Av. OCC/min
T-0.5H
REF 14 3 6 7.7
SOL 4 3 6 6.5
T-12H
REF 29 4 8 5
SOL 4 3 6 5.5

At T-0.5H, the REF scenario changes to CONF.14 while Figure 15. CC distribution for CONF.29 at T-12H
SOL scenario gives CONF.4 as the most suitable. When
 o Capacity: improvement in suitability and
predictability;
o Cost-efficiency: improvement in ATCo control
hours and workload.
The recommendations for the next research phases are:
 The stability of the configuration selection along time
must be ensured to avoid unnecessary changes with no
significant increase in performance in very short-term.
 Weighting of the parameters feeding the sector
configuration optimisation can be further researched as
dependent of the ANSP strategy.
Figure 16. CC distribution for CONF.4 at T-12H  The visualisation of potential benefits of each
configuration should be provided to help decision-
As a summary, the benefits in Cost-efficiency can be proven making.
since there was in SOL scenarios a reduction in number of
sectors, in average OCC/min or in a more balanced workload As a general conclusion, the proposed sector configuration
distribution. optimiser for the Enhanced DCB management in DAC concept
provides a clear potential for the airspace capacity improvement
V. CONCLUSIONS and further research is recommended for this topic to complete
the work and implement the solution.
Based on results presented, it can be generally stated that
there are significant benefits expected from the application of
the Enhanced Complexity Assessment. A clear evidence of its ACKNOWLEDGMENT
technical and operational feasibility is provided with the The authors want to dedicate a special mention to Shutao
following summary of work done and operational benefits: Zheng, former CRIDA R&D researcher, who dedicated a huge
effort on the development of the DAC optimizer algorithm
 Complexity assessment to the DAC concept for DCB prototype.
process enhancement:
o Cognitive complexity (CC); REFERENCES
o Sector Shape complexity (SS); [1] SESAR 2020 PJ08 Advanced Airspace Management. SESAR Solution
08.01 SPR-INTEROP-OSED for V2 – Part I. Edition 03.00.01, SESAR
o Uncertainty assessment (PCC). Joint Undertaking, 2019.
[2] SESAR 2020 PJ08 Advanced Airspace Management. SESAR Solution
 Sector configuration Optimiser tool: 08.01 Validation Report (VALR) for V2. Edition 00.03.01, SESAR Joint
Undertaking, 2019.
o Cost function definition taking into account:
[3] L. Basora, V. Courchelle, J. Bedouet, and T. Dubot, ‘Occupancy Peak
 Overloads and underloads; Estimation from Sector Geometry and Traffic Flow Data’, Proc 8th
SESAR Innov. Days, 2018.
 PCC threshold and weights calibration; [4] P. López de Frutos, R. Rodríguez Rodríguez, D. Zheng Zhang, S. Zheng,
J. J. Cañas, E. Muñoz de Escalona, “COMETA: An Air Traffic
 Average and balance between sectors. Controller’s Mental Workload Model for Calculating and Predicting
Demand and Capacity Balancing”, H-WORKLOAD 2019.
o Applied to enhanced DAC DCB process.
[5] Jordan and Brennen, “Instantaneous self-assessment of workload
 Capacity and cost-efficiency improvement assessment, technique (ISA)” 1992.
resulted in positive results: